first commit
This commit is contained in:
113
dev/online_quant_configs/DeepSeek-R1-0528-w4a8/config.json
Normal file
113
dev/online_quant_configs/DeepSeek-R1-0528-w4a8/config.json
Normal file
@@ -0,0 +1,113 @@
|
||||
{
|
||||
"architectures": [
|
||||
"DeepseekV3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_deepseek.DeepseekV3Config",
|
||||
"AutoModel": "modeling_deepseek.DeepseekV3Model",
|
||||
"AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
|
||||
},
|
||||
"bos_token_id": 0,
|
||||
"eos_token_id": 1,
|
||||
"ep_size": 1,
|
||||
"first_k_dense_replace": 3,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 7168,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 18432,
|
||||
"kv_lora_rank": 512,
|
||||
"max_position_embeddings": 163840,
|
||||
"model_type": "deepseek_v3",
|
||||
"moe_intermediate_size": 2048,
|
||||
"moe_layer_freq": 1,
|
||||
"n_group": 8,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": 1,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 128,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 61,
|
||||
"num_key_value_heads": 128,
|
||||
"num_nextn_predict_layers": 1,
|
||||
"q_lora_rank": 1536,
|
||||
"qk_nope_head_dim": 128,
|
||||
"qk_rope_head_dim": 64,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": {
|
||||
"beta_fast": 32,
|
||||
"beta_slow": 1,
|
||||
"factor": 40,
|
||||
"mscale": 1.0,
|
||||
"mscale_all_dim": 1.0,
|
||||
"original_max_position_embeddings": 4096,
|
||||
"type": "yarn"
|
||||
},
|
||||
"rope_theta": 10000,
|
||||
"routed_scaling_factor": 2.5,
|
||||
"scoring_func": "sigmoid",
|
||||
"tie_word_embeddings": false,
|
||||
"topk_group": 4,
|
||||
"topk_method": "noaux_tc",
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.46.3",
|
||||
"use_cache": true,
|
||||
"v_head_dim": 128,
|
||||
"vocab_size": 129280,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*input_layernorm.*",
|
||||
"re:.*post_attention_layernorm.*",
|
||||
"re:.*kv_a_layernorm.*",
|
||||
"re:.*q_a_layernorm.*",
|
||||
"re:.*eh_proj.*",
|
||||
"re:.*enorm.*",
|
||||
"re:.*hnorm.*",
|
||||
"re:model.layers.\\d+.mlp.gate_proj.*",
|
||||
"re:model.layers.\\d+.mlp.up_proj.*",
|
||||
"re:model.layers.\\d+.mlp.down_proj.*",
|
||||
"re:model.layers.\\d+.mlp.gate\\..*",
|
||||
"re:model.norm.*",
|
||||
"re:model.layers.\\d+.shared_head.norm.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:model.layers.\\d+.shared_head.head.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,199 @@
|
||||
from transformers.configuration_utils import PretrainedConfig
|
||||
from transformers.utils import logging
|
||||
|
||||
logger = logging.get_logger(__name__)
|
||||
|
||||
DEEPSEEK_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
|
||||
class DeepseekV3Config(PretrainedConfig):
|
||||
r"""
|
||||
This is the configuration class to store the configuration of a [`DeepseekV3Model`]. It is used to instantiate an DeepSeek
|
||||
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
|
||||
defaults will yield a similar configuration to that of the DeepSeek-V3.
|
||||
|
||||
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
||||
documentation from [`PretrainedConfig`] for more information.
|
||||
|
||||
|
||||
Args:
|
||||
vocab_size (`int`, *optional*, defaults to 129280):
|
||||
Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
|
||||
`inputs_ids` passed when calling [`DeepseekV3Model`]
|
||||
hidden_size (`int`, *optional*, defaults to 4096):
|
||||
Dimension of the hidden representations.
|
||||
intermediate_size (`int`, *optional*, defaults to 11008):
|
||||
Dimension of the MLP representations.
|
||||
moe_intermediate_size (`int`, *optional*, defaults to 1407):
|
||||
Dimension of the MoE representations.
|
||||
num_hidden_layers (`int`, *optional*, defaults to 32):
|
||||
Number of hidden layers in the Transformer decoder.
|
||||
num_nextn_predict_layers (`int`, *optional*, defaults to 1):
|
||||
Number of nextn predict layers in the DeepSeekV3 Model.
|
||||
num_attention_heads (`int`, *optional*, defaults to 32):
|
||||
Number of attention heads for each attention layer in the Transformer decoder.
|
||||
n_shared_experts (`int`, *optional*, defaults to None):
|
||||
Number of shared experts, None means dense model.
|
||||
n_routed_experts (`int`, *optional*, defaults to None):
|
||||
Number of routed experts, None means dense model.
|
||||
routed_scaling_factor (`float`, *optional*, defaults to 1.0):
|
||||
Scaling factor or routed experts.
|
||||
topk_method (`str`, *optional*, defaults to `gready`):
|
||||
Topk method used in routed gate.
|
||||
n_group (`int`, *optional*, defaults to None):
|
||||
Number of groups for routed experts.
|
||||
topk_group (`int`, *optional*, defaults to None):
|
||||
Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
|
||||
num_experts_per_tok (`int`, *optional*, defaults to None):
|
||||
Number of selected experts, None means dense model.
|
||||
moe_layer_freq (`int`, *optional*, defaults to 1):
|
||||
The frequency of the MoE layer: one expert layer for every `moe_layer_freq - 1` dense layers.
|
||||
first_k_dense_replace (`int`, *optional*, defaults to 0):
|
||||
Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
|
||||
\--k dense layers--/
|
||||
norm_topk_prob (`bool`, *optional*, defaults to False):
|
||||
Whether to normalize the weights of the routed experts.
|
||||
scoring_func (`str`, *optional*, defaults to 'softmax'):
|
||||
Method of computing expert weights.
|
||||
aux_loss_alpha (`float`, *optional*, defaults to 0.001):
|
||||
Auxiliary loss weight coefficient.
|
||||
seq_aux = (`bool`, *optional*, defaults to True):
|
||||
Whether to compute the auxiliary loss for each individual sample.
|
||||
num_key_value_heads (`int`, *optional*):
|
||||
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
|
||||
`num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
|
||||
`num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
|
||||
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
|
||||
by meanpooling all the original heads within that group. For more details checkout [this
|
||||
paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
|
||||
`num_attention_heads`.
|
||||
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
|
||||
The non-linear activation function (function or string) in the decoder.
|
||||
max_position_embeddings (`int`, *optional*, defaults to 2048):
|
||||
The maximum sequence length that this model might ever be used with.
|
||||
initializer_range (`float`, *optional*, defaults to 0.02):
|
||||
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
||||
rms_norm_eps (`float`, *optional*, defaults to 1e-06):
|
||||
The epsilon used by the rms normalization layers.
|
||||
use_cache (`bool`, *optional*, defaults to `True`):
|
||||
Whether or not the model should return the last key/values attentions (not used by all models). Only
|
||||
relevant if `config.is_decoder=True`.
|
||||
pad_token_id (`int`, *optional*):
|
||||
Padding token id.
|
||||
bos_token_id (`int`, *optional*, defaults to 1):
|
||||
Beginning of stream token id.
|
||||
eos_token_id (`int`, *optional*, defaults to 2):
|
||||
End of stream token id.
|
||||
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
|
||||
Whether to tie weight embeddings
|
||||
rope_theta (`float`, *optional*, defaults to 10000.0):
|
||||
The base period of the RoPE embeddings.
|
||||
rope_scaling (`Dict`, *optional*):
|
||||
Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
|
||||
strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
|
||||
`{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
|
||||
`max_position_embeddings` to the expected new maximum.
|
||||
attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
|
||||
Whether to use a bias in the query, key, value and output projection layers during self-attention.
|
||||
attention_dropout (`float`, *optional*, defaults to 0.0):
|
||||
The dropout ratio for the attention probabilities.
|
||||
|
||||
```python
|
||||
>>> from transformers import DeepseekV3Model, DeepseekV3Config
|
||||
|
||||
>>> # Initializing a Deepseek-V3 style configuration
|
||||
>>> configuration = DeepseekV3Config()
|
||||
|
||||
>>> # Accessing the model configuration
|
||||
>>> configuration = model.config
|
||||
```"""
|
||||
|
||||
model_type = "deepseek_v3"
|
||||
keys_to_ignore_at_inference = ["past_key_values"]
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
vocab_size=129280,
|
||||
hidden_size=7168,
|
||||
intermediate_size=18432,
|
||||
moe_intermediate_size = 2048,
|
||||
num_hidden_layers=61,
|
||||
num_nextn_predict_layers=1,
|
||||
num_attention_heads=128,
|
||||
num_key_value_heads=128,
|
||||
n_shared_experts = 1,
|
||||
n_routed_experts = 256,
|
||||
ep_size = 1,
|
||||
routed_scaling_factor = 2.5,
|
||||
kv_lora_rank = 512,
|
||||
q_lora_rank = 1536,
|
||||
qk_rope_head_dim = 64,
|
||||
v_head_dim = 128,
|
||||
qk_nope_head_dim = 128,
|
||||
topk_method = 'noaux_tc',
|
||||
n_group = 8,
|
||||
topk_group = 4,
|
||||
num_experts_per_tok = 8,
|
||||
moe_layer_freq = 1,
|
||||
first_k_dense_replace = 3,
|
||||
norm_topk_prob = True,
|
||||
scoring_func = 'sigmoid',
|
||||
hidden_act="silu",
|
||||
max_position_embeddings=4096,
|
||||
initializer_range=0.02,
|
||||
rms_norm_eps=1e-6,
|
||||
use_cache=True,
|
||||
pad_token_id=None,
|
||||
bos_token_id=0,
|
||||
eos_token_id=1,
|
||||
tie_word_embeddings=False,
|
||||
rope_theta=10000.0,
|
||||
rope_scaling=None,
|
||||
attention_bias=False,
|
||||
attention_dropout=0.0,
|
||||
**kwargs,
|
||||
):
|
||||
self.vocab_size = vocab_size
|
||||
self.max_position_embeddings = max_position_embeddings
|
||||
self.hidden_size = hidden_size
|
||||
self.intermediate_size = intermediate_size
|
||||
self.moe_intermediate_size = moe_intermediate_size
|
||||
self.num_hidden_layers = num_hidden_layers
|
||||
self.num_nextn_predict_layers = num_nextn_predict_layers
|
||||
self.num_attention_heads = num_attention_heads
|
||||
self.n_shared_experts = n_shared_experts
|
||||
self.n_routed_experts = n_routed_experts
|
||||
self.ep_size = ep_size
|
||||
self.routed_scaling_factor = routed_scaling_factor
|
||||
self.kv_lora_rank = kv_lora_rank
|
||||
self.q_lora_rank = q_lora_rank
|
||||
self.qk_rope_head_dim = qk_rope_head_dim
|
||||
self.v_head_dim = v_head_dim
|
||||
self.qk_nope_head_dim = qk_nope_head_dim
|
||||
self.topk_method = topk_method
|
||||
self.n_group = n_group
|
||||
self.topk_group = topk_group
|
||||
self.num_experts_per_tok = num_experts_per_tok
|
||||
self.moe_layer_freq = moe_layer_freq
|
||||
self.first_k_dense_replace = first_k_dense_replace
|
||||
self.norm_topk_prob = norm_topk_prob
|
||||
self.scoring_func = scoring_func
|
||||
# for backward compatibility
|
||||
if num_key_value_heads is None:
|
||||
num_key_value_heads = num_attention_heads
|
||||
|
||||
self.num_key_value_heads = num_key_value_heads
|
||||
self.hidden_act = hidden_act
|
||||
self.initializer_range = initializer_range
|
||||
self.rms_norm_eps = rms_norm_eps
|
||||
self.use_cache = use_cache
|
||||
self.rope_theta = rope_theta
|
||||
self.rope_scaling = rope_scaling
|
||||
self.attention_bias = attention_bias
|
||||
self.attention_dropout = attention_dropout
|
||||
|
||||
super().__init__(
|
||||
pad_token_id=pad_token_id,
|
||||
bos_token_id=bos_token_id,
|
||||
eos_token_id=eos_token_id,
|
||||
tie_word_embeddings=tie_word_embeddings,
|
||||
**kwargs,
|
||||
)
|
||||
46185
dev/online_quant_configs/DeepSeek-R1-0528-w4a8/quant_map.json
Normal file
46185
dev/online_quant_configs/DeepSeek-R1-0528-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
58
dev/online_quant_configs/DeepSeek-R1-0528/config.json
Normal file
58
dev/online_quant_configs/DeepSeek-R1-0528/config.json
Normal file
@@ -0,0 +1,58 @@
|
||||
{
|
||||
"architectures": [
|
||||
"DeepseekV3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_deepseek.DeepseekV3Config",
|
||||
"AutoModel": "modeling_deepseek.DeepseekV3Model",
|
||||
"AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
|
||||
},
|
||||
"bos_token_id": 0,
|
||||
"eos_token_id": 1,
|
||||
"ep_size": 1,
|
||||
"first_k_dense_replace": 3,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 7168,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 18432,
|
||||
"kv_lora_rank": 512,
|
||||
"max_position_embeddings": 163840,
|
||||
"model_type": "deepseek_v3",
|
||||
"moe_intermediate_size": 2048,
|
||||
"moe_layer_freq": 1,
|
||||
"n_group": 8,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": 1,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 128,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 61,
|
||||
"num_key_value_heads": 128,
|
||||
"num_nextn_predict_layers": 1,
|
||||
"q_lora_rank": 1536,
|
||||
"qk_nope_head_dim": 128,
|
||||
"qk_rope_head_dim": 64,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": {
|
||||
"beta_fast": 32,
|
||||
"beta_slow": 1,
|
||||
"factor": 40,
|
||||
"mscale": 1.0,
|
||||
"mscale_all_dim": 1.0,
|
||||
"original_max_position_embeddings": 4096,
|
||||
"type": "yarn"
|
||||
},
|
||||
"rope_theta": 10000,
|
||||
"routed_scaling_factor": 2.5,
|
||||
"scoring_func": "sigmoid",
|
||||
"tie_word_embeddings": false,
|
||||
"topk_group": 4,
|
||||
"topk_method": "noaux_tc",
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.46.3",
|
||||
"use_cache": true,
|
||||
"v_head_dim": 128,
|
||||
"vocab_size": 129280
|
||||
}
|
||||
110
dev/online_quant_configs/GLM-5-w4a8-nn-g64/config.json
Normal file
110
dev/online_quant_configs/GLM-5-w4a8-nn-g64/config.json
Normal file
@@ -0,0 +1,110 @@
|
||||
{
|
||||
"architectures": [
|
||||
"GlmMoeDsaForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": [
|
||||
154820,
|
||||
154827,
|
||||
154829
|
||||
],
|
||||
"ep_size": 1,
|
||||
"first_k_dense_replace": 3,
|
||||
"hidden_act": "silu",
|
||||
"head_dim": 64,
|
||||
"hidden_size": 6144,
|
||||
"index_head_dim": 128,
|
||||
"index_n_heads": 32,
|
||||
"index_topk": 2048,
|
||||
"indexer_rope_interleave": true,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 12288,
|
||||
"kv_lora_rank": 512,
|
||||
"max_position_embeddings": 202752,
|
||||
"moe_intermediate_size": 2048,
|
||||
"moe_layer_freq": 1,
|
||||
"model_type": "glm_moe_dsa",
|
||||
"n_group": 1,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": 1,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 64,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 78,
|
||||
"num_key_value_heads": 64,
|
||||
"num_nextn_predict_layers": 1,
|
||||
"pad_token_id": 154820,
|
||||
"pretraining_tp": 1,
|
||||
"q_lora_rank": 2048,
|
||||
"qk_head_dim": 256,
|
||||
"qk_nope_head_dim": 192,
|
||||
"qk_rope_head_dim": 64,
|
||||
"rms_norm_eps": 1e-05,
|
||||
"rope_interleave": true,
|
||||
"rope_parameters": {
|
||||
"rope_theta": 1000000,
|
||||
"rope_type": "default"
|
||||
},
|
||||
"routed_scaling_factor": 2.5,
|
||||
"scoring_func": "sigmoid",
|
||||
"tie_word_embeddings": false,
|
||||
"topk_group": 1,
|
||||
"topk_method": "noaux_tc",
|
||||
"transformers_version": "5.0.2.dev0",
|
||||
"use_cache": true,
|
||||
"v_head_dim": 256,
|
||||
"vocab_size": 154880,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": 64,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "group",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*self_attn.*",
|
||||
"re:.*bias.*",
|
||||
"re:.*mlp.gate.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:.*eh_proj.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*shared_head.*",
|
||||
"re:model.layers.\\d+.mlp.shared_experts.*",
|
||||
"re:model.layers.\\d+.mlp.down_proj.*",
|
||||
"re:model.layers.\\d+.mlp.up_proj.*",
|
||||
"re:model.layers.\\d+.mlp.proj_proj.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
59872
dev/online_quant_configs/GLM-5-w4a8-nn-g64/quant_map.json
Normal file
59872
dev/online_quant_configs/GLM-5-w4a8-nn-g64/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
110
dev/online_quant_configs/GLM-5-w4a8/config.json
Normal file
110
dev/online_quant_configs/GLM-5-w4a8/config.json
Normal file
@@ -0,0 +1,110 @@
|
||||
{
|
||||
"architectures": [
|
||||
"GlmMoeDsaForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": [
|
||||
154820,
|
||||
154827,
|
||||
154829
|
||||
],
|
||||
"ep_size": 1,
|
||||
"first_k_dense_replace": 3,
|
||||
"hidden_act": "silu",
|
||||
"head_dim": 64,
|
||||
"hidden_size": 6144,
|
||||
"index_head_dim": 128,
|
||||
"index_n_heads": 32,
|
||||
"index_topk": 2048,
|
||||
"indexer_rope_interleave": true,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 12288,
|
||||
"kv_lora_rank": 512,
|
||||
"max_position_embeddings": 202752,
|
||||
"moe_intermediate_size": 2048,
|
||||
"moe_layer_freq": 1,
|
||||
"model_type": "glm_moe_dsa",
|
||||
"n_group": 1,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": 1,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 64,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 78,
|
||||
"num_key_value_heads": 64,
|
||||
"num_nextn_predict_layers": 1,
|
||||
"pad_token_id": 154820,
|
||||
"pretraining_tp": 1,
|
||||
"q_lora_rank": 2048,
|
||||
"qk_head_dim": 256,
|
||||
"qk_nope_head_dim": 192,
|
||||
"qk_rope_head_dim": 64,
|
||||
"rms_norm_eps": 1e-05,
|
||||
"rope_interleave": true,
|
||||
"rope_parameters": {
|
||||
"rope_theta": 1000000,
|
||||
"rope_type": "default"
|
||||
},
|
||||
"routed_scaling_factor": 2.5,
|
||||
"scoring_func": "sigmoid",
|
||||
"tie_word_embeddings": false,
|
||||
"topk_group": 1,
|
||||
"topk_method": "noaux_tc",
|
||||
"transformers_version": "5.0.2.dev0",
|
||||
"use_cache": true,
|
||||
"v_head_dim": 256,
|
||||
"vocab_size": 154880,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*self_attn.*",
|
||||
"re:.*bias.*",
|
||||
"re:.*mlp.gate.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:.*eh_proj.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*shared_head.*",
|
||||
"re:model.layers.\\d+.mlp.shared_experts.*",
|
||||
"re:model.layers.\\d+.mlp.down_proj.*",
|
||||
"re:model.layers.\\d+.mlp.up_proj.*",
|
||||
"re:model.layers.\\d+.mlp.proj_proj.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
59872
dev/online_quant_configs/GLM-5-w4a8/quant_map.json
Normal file
59872
dev/online_quant_configs/GLM-5-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
197
dev/online_quant_configs/MiMo-V2-Flash-w4a8/config.json
Normal file
197
dev/online_quant_configs/MiMo-V2-Flash-w4a8/config.json
Normal file
@@ -0,0 +1,197 @@
|
||||
{
|
||||
"architectures": [
|
||||
"MiMoV2FlashForCausalLM"
|
||||
],
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_mimo_v2_flash.MiMoV2FlashConfig",
|
||||
"AutoModel": "modeling_mimo_v2_flash.MiMoV2FlashModel",
|
||||
"AutoModelForCausalLM": "modeling_mimo_v2_flash.MiMoV2FlashForCausalLM"
|
||||
},
|
||||
"attention_dropout": 0.0,
|
||||
"attention_value_scale": 0.707,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 4096,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 16384,
|
||||
"max_position_embeddings": 262144,
|
||||
"model_type": "mimo_v2_flash",
|
||||
"num_attention_heads": 64,
|
||||
"head_dim": 192,
|
||||
"num_hidden_layers": 48,
|
||||
"num_key_value_heads": 4,
|
||||
"layernorm_epsilon": 1e-05,
|
||||
"rope_theta": 5000000,
|
||||
"tie_word_embeddings": false,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.40.1",
|
||||
"use_cache": true,
|
||||
"vocab_size": 152576,
|
||||
"partial_rotary_factor": 0.334,
|
||||
"sliding_window": 128,
|
||||
"swa_rope_theta": 10000,
|
||||
"attention_bias": false,
|
||||
"v_head_dim": 128,
|
||||
"hybrid_layer_pattern": [
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0
|
||||
],
|
||||
"add_swa_attention_sink_bias": true,
|
||||
"add_full_attention_sink_bias": false,
|
||||
"sliding_window_size": 128,
|
||||
"attention_chunk_size": 128,
|
||||
"moe_layer_freq": [
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1
|
||||
],
|
||||
"moe_intermediate_size": 2048,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": null,
|
||||
"num_experts_per_tok": 8,
|
||||
"norm_topk_prob": true,
|
||||
"scoring_func": "sigmoid",
|
||||
"n_group": 1,
|
||||
"topk_group": 1,
|
||||
"topk_method": "noaux_tc",
|
||||
"routed_scaling_factor": null,
|
||||
"swa_num_attention_heads": 64,
|
||||
"swa_num_key_value_heads": 8,
|
||||
"swa_head_dim": 192,
|
||||
"swa_v_head_dim": 128,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:model.layers.\\d+.mlp.gate\\..*",
|
||||
"re:.*bias.*",
|
||||
"re:.*.mtp.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
36567
dev/online_quant_configs/MiMo-V2-Flash-w4a8/quant_map.json
Normal file
36567
dev/online_quant_configs/MiMo-V2-Flash-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
197
dev/online_quant_configs/MiMo-V2-Flash-w8a8/config.json
Normal file
197
dev/online_quant_configs/MiMo-V2-Flash-w8a8/config.json
Normal file
@@ -0,0 +1,197 @@
|
||||
{
|
||||
"architectures": [
|
||||
"MiMoV2FlashForCausalLM"
|
||||
],
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_mimo_v2_flash.MiMoV2FlashConfig",
|
||||
"AutoModel": "modeling_mimo_v2_flash.MiMoV2FlashModel",
|
||||
"AutoModelForCausalLM": "modeling_mimo_v2_flash.MiMoV2FlashForCausalLM"
|
||||
},
|
||||
"attention_dropout": 0.0,
|
||||
"attention_value_scale": 0.707,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 4096,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 16384,
|
||||
"max_position_embeddings": 262144,
|
||||
"model_type": "mimo_v2_flash",
|
||||
"num_attention_heads": 64,
|
||||
"head_dim": 192,
|
||||
"num_hidden_layers": 48,
|
||||
"num_key_value_heads": 4,
|
||||
"layernorm_epsilon": 1e-05,
|
||||
"rope_theta": 5000000,
|
||||
"tie_word_embeddings": false,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.40.1",
|
||||
"use_cache": true,
|
||||
"vocab_size": 152576,
|
||||
"partial_rotary_factor": 0.334,
|
||||
"sliding_window": 128,
|
||||
"swa_rope_theta": 10000,
|
||||
"attention_bias": false,
|
||||
"v_head_dim": 128,
|
||||
"hybrid_layer_pattern": [
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
0
|
||||
],
|
||||
"add_swa_attention_sink_bias": true,
|
||||
"add_full_attention_sink_bias": false,
|
||||
"sliding_window_size": 128,
|
||||
"attention_chunk_size": 128,
|
||||
"moe_layer_freq": [
|
||||
0,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1
|
||||
],
|
||||
"moe_intermediate_size": 2048,
|
||||
"n_routed_experts": 256,
|
||||
"n_shared_experts": null,
|
||||
"num_experts_per_tok": 8,
|
||||
"norm_topk_prob": true,
|
||||
"scoring_func": "sigmoid",
|
||||
"n_group": 1,
|
||||
"topk_group": 1,
|
||||
"topk_method": "noaux_tc",
|
||||
"routed_scaling_factor": null,
|
||||
"swa_num_attention_heads": 64,
|
||||
"swa_num_key_value_heads": 8,
|
||||
"swa_head_dim": 192,
|
||||
"swa_v_head_dim": 128,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:model.layers.\\d+.mlp.gate\\..*",
|
||||
"re:.*bias.*",
|
||||
"re:.*.mtp.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
36567
dev/online_quant_configs/MiMo-V2-Flash-w8a8/quant_map.json
Normal file
36567
dev/online_quant_configs/MiMo-V2-Flash-w8a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
146
dev/online_quant_configs/MiniMax-M2.5-w4a8/config.json
Normal file
146
dev/online_quant_configs/MiniMax-M2.5-w4a8/config.json
Normal file
@@ -0,0 +1,146 @@
|
||||
{
|
||||
"architectures": [
|
||||
"MiniMaxM2ForCausalLM"
|
||||
],
|
||||
"attn_type_list": [
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1
|
||||
],
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_minimax_m2.MiniMaxM2Config",
|
||||
"AutoModelForCausalLM": "modeling_minimax_m2.MiniMaxM2ForCausalLM"
|
||||
},
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 3072,
|
||||
"intermediate_size": 1536,
|
||||
"max_position_embeddings": 196608,
|
||||
"model_type": "minimax_m2",
|
||||
"mtp_transformer_layers": 1,
|
||||
"num_attention_heads": 48,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 62,
|
||||
"num_key_value_heads": 8,
|
||||
"num_local_experts": 256,
|
||||
"num_mtp_modules": 3,
|
||||
"qk_norm_type": "per_layer",
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_theta": 5000000,
|
||||
"rotary_dim": 64,
|
||||
"scoring_func": "sigmoid",
|
||||
"shared_intermediate_size": 0,
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "4.46.1",
|
||||
"use_cache": true,
|
||||
"use_mtp": true,
|
||||
"use_qk_norm": true,
|
||||
"use_routing_bias": true,
|
||||
"vocab_size": 200064,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"model.embed_tokens.weight",
|
||||
"re:model.layers.\\d+.input_layernorm.weight",
|
||||
"re:model.layers.\\d+.self_attn.*",
|
||||
"re:model.layers.\\d+.post_attention_layernorm.weight",
|
||||
"re:model.layers.\\d+.block_sparse_moe.gate.weight",
|
||||
"re:model.layers.\\d+.block_sparse_moe.e_score_correction_bias",
|
||||
"model.norm.weight",
|
||||
"lm_head.weight"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
48241
dev/online_quant_configs/MiniMax-M2.5-w4a8/quant_map.json
Normal file
48241
dev/online_quant_configs/MiniMax-M2.5-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,85 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3MoeForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"decoder_sparse_step": 1,
|
||||
"eos_token_id": 151645,
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2048,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 6144,
|
||||
"max_position_embeddings": 262144,
|
||||
"max_window_layers": 48,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_moe",
|
||||
"moe_intermediate_size": 768,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 32,
|
||||
"num_experts": 128,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 48,
|
||||
"num_key_value_heads": 4,
|
||||
"output_router_logits": false,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": null,
|
||||
"rope_theta": 10000000,
|
||||
"router_aux_loss_coef": 0.001,
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": false,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.51.0",
|
||||
"use_cache": true,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:model.norm.*",
|
||||
"re:model.embed_tokens.*",
|
||||
"re:model.layers.\\d+.input_layernorm.*",
|
||||
"re:model.layers.\\d+.mlp.gate.*",
|
||||
"re:model.layers.\\d+.post_attention_layernorm.*",
|
||||
"re:model.layers.\\d+.self_attn.k_norm.*",
|
||||
"re:model.layers.\\d+.self_attn.q_norm.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
18869
dev/online_quant_configs/Qwen3-30B-A3B-Instruct-2507-w4a8/quant_map.json
Normal file
18869
dev/online_quant_configs/Qwen3-30B-A3B-Instruct-2507-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,95 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3NextForCausalLM"
|
||||
],
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"decoder_sparse_step": 1,
|
||||
"eos_token_id": 151645,
|
||||
"full_attention_interval": 4,
|
||||
"head_dim": 256,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2048,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 5120,
|
||||
"linear_conv_kernel_dim": 4,
|
||||
"linear_key_head_dim": 128,
|
||||
"linear_num_key_heads": 16,
|
||||
"linear_num_value_heads": 32,
|
||||
"linear_value_head_dim": 128,
|
||||
"max_position_embeddings": 262144,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_next",
|
||||
"moe_intermediate_size": 512,
|
||||
"norm_topk_prob": true,
|
||||
"num_attention_heads": 16,
|
||||
"num_experts": 512,
|
||||
"num_experts_per_tok": 10,
|
||||
"num_hidden_layers": 48,
|
||||
"num_key_value_heads": 2,
|
||||
"output_router_logits": false,
|
||||
"partial_rotary_factor": 0.25,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": null,
|
||||
"rope_theta": 10000000,
|
||||
"router_aux_loss_coef": 0.001,
|
||||
"shared_expert_intermediate_size": 512,
|
||||
"tie_word_embeddings": false,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.57.0.dev0",
|
||||
"use_cache": true,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*linear_attn.*",
|
||||
"re:.*self_attn.*",
|
||||
"re:.*shared_expert.*",
|
||||
"re:.*linear_attn.in_proj_ba.*",
|
||||
"re:.*conv1d.*",
|
||||
"re:.*A_log.*",
|
||||
"re:.*shared_expert_gate.*",
|
||||
"re:.*bias.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*mlp.gate.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:.*mtp.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
75946
dev/online_quant_configs/Qwen3-Next-80B-A3B-Thinking-w4a8/quant_map.json
Normal file
75946
dev/online_quant_configs/Qwen3-Next-80B-A3B-Thinking-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
180
dev/online_quant_configs/Qwen3.5-27B-w4a8/config.json
Normal file
180
dev/online_quant_configs/Qwen3.5-27B-w4a8/config.json
Normal file
@@ -0,0 +1,180 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3_5ForConditionalGeneration"
|
||||
],
|
||||
"image_token_id": 248056,
|
||||
"model_type": "qwen3_5",
|
||||
"text_config": {
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"attn_output_gate": true,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": 248044,
|
||||
"full_attention_interval": 4,
|
||||
"head_dim": 256,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 5120,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 17408,
|
||||
"layer_types": [
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"linear_conv_kernel_dim": 4,
|
||||
"linear_key_head_dim": 128,
|
||||
"linear_num_key_heads": 16,
|
||||
"linear_num_value_heads": 48,
|
||||
"linear_value_head_dim": 128,
|
||||
"max_position_embeddings": 262144,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_5_text",
|
||||
"mtp_num_hidden_layers": 1,
|
||||
"mtp_use_dedicated_embeddings": false,
|
||||
"num_attention_heads": 24,
|
||||
"num_hidden_layers": 64,
|
||||
"num_key_value_heads": 4,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"use_cache": true,
|
||||
"vocab_size": 248320,
|
||||
"mamba_ssm_dtype": "float32",
|
||||
"rope_parameters": {
|
||||
"mrope_interleaved": true,
|
||||
"mrope_section": [
|
||||
11,
|
||||
11,
|
||||
10
|
||||
],
|
||||
"rope_type": "default",
|
||||
"rope_theta": 10000000,
|
||||
"partial_rotary_factor": 0.25
|
||||
}
|
||||
},
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "4.57.0.dev0",
|
||||
"video_token_id": 248057,
|
||||
"vision_config": {
|
||||
"deepstack_visual_indexes": [],
|
||||
"depth": 27,
|
||||
"hidden_act": "gelu_pytorch_tanh",
|
||||
"hidden_size": 1152,
|
||||
"in_channels": 3,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "qwen3_5",
|
||||
"num_heads": 16,
|
||||
"num_position_embeddings": 2304,
|
||||
"out_hidden_size": 5120,
|
||||
"patch_size": 16,
|
||||
"spatial_merge_size": 2,
|
||||
"temporal_patch_size": 2
|
||||
},
|
||||
"vision_end_token_id": 248054,
|
||||
"vision_start_token_id": 248053,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*visual.*",
|
||||
"re:.*mtp.*",
|
||||
"re:.*linear_attn.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*embed_tokens.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
1201
dev/online_quant_configs/Qwen3.5-27B-w4a8/quant_map.json
Normal file
1201
dev/online_quant_configs/Qwen3.5-27B-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
180
dev/online_quant_configs/Qwen3.5-27B-w8a8/config.json
Normal file
180
dev/online_quant_configs/Qwen3.5-27B-w8a8/config.json
Normal file
@@ -0,0 +1,180 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3_5ForConditionalGeneration"
|
||||
],
|
||||
"image_token_id": 248056,
|
||||
"model_type": "qwen3_5",
|
||||
"text_config": {
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"attn_output_gate": true,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": 248044,
|
||||
"full_attention_interval": 4,
|
||||
"head_dim": 256,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 5120,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 17408,
|
||||
"layer_types": [
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"linear_conv_kernel_dim": 4,
|
||||
"linear_key_head_dim": 128,
|
||||
"linear_num_key_heads": 16,
|
||||
"linear_num_value_heads": 48,
|
||||
"linear_value_head_dim": 128,
|
||||
"max_position_embeddings": 262144,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_5_text",
|
||||
"mtp_num_hidden_layers": 1,
|
||||
"mtp_use_dedicated_embeddings": false,
|
||||
"num_attention_heads": 24,
|
||||
"num_hidden_layers": 64,
|
||||
"num_key_value_heads": 4,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"use_cache": true,
|
||||
"vocab_size": 248320,
|
||||
"mamba_ssm_dtype": "float32",
|
||||
"rope_parameters": {
|
||||
"mrope_interleaved": true,
|
||||
"mrope_section": [
|
||||
11,
|
||||
11,
|
||||
10
|
||||
],
|
||||
"rope_type": "default",
|
||||
"rope_theta": 10000000,
|
||||
"partial_rotary_factor": 0.25
|
||||
}
|
||||
},
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "4.57.0.dev0",
|
||||
"video_token_id": 248057,
|
||||
"vision_config": {
|
||||
"deepstack_visual_indexes": [],
|
||||
"depth": 27,
|
||||
"hidden_act": "gelu_pytorch_tanh",
|
||||
"hidden_size": 1152,
|
||||
"in_channels": 3,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "qwen3_5",
|
||||
"num_heads": 16,
|
||||
"num_position_embeddings": 2304,
|
||||
"out_hidden_size": 5120,
|
||||
"patch_size": 16,
|
||||
"spatial_merge_size": 2,
|
||||
"temporal_patch_size": 2
|
||||
},
|
||||
"vision_end_token_id": 248054,
|
||||
"vision_start_token_id": 248053,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*visual.*",
|
||||
"re:.*mtp.*",
|
||||
"re:.*linear_attn.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*embed_tokens.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
1201
dev/online_quant_configs/Qwen3.5-27B-w8a8/quant_map.json
Normal file
1201
dev/online_quant_configs/Qwen3.5-27B-w8a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
162
dev/online_quant_configs/Qwen3.5-35B-A3B-w4a8/config.json
Normal file
162
dev/online_quant_configs/Qwen3.5-35B-A3B-w4a8/config.json
Normal file
@@ -0,0 +1,162 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3_5MoeForConditionalGeneration"
|
||||
],
|
||||
"image_token_id": 248056,
|
||||
"model_type": "qwen3_5_moe",
|
||||
"text_config": {
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"attn_output_gate": true,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": 248044,
|
||||
"full_attention_interval": 4,
|
||||
"head_dim": 256,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2048,
|
||||
"initializer_range": 0.02,
|
||||
"layer_types": [
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"linear_conv_kernel_dim": 4,
|
||||
"linear_key_head_dim": 128,
|
||||
"linear_num_key_heads": 16,
|
||||
"linear_num_value_heads": 32,
|
||||
"linear_value_head_dim": 128,
|
||||
"max_position_embeddings": 262144,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_5_moe_text",
|
||||
"moe_intermediate_size": 512,
|
||||
"mtp_num_hidden_layers": 1,
|
||||
"mtp_use_dedicated_embeddings": false,
|
||||
"num_attention_heads": 16,
|
||||
"num_experts": 256,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 40,
|
||||
"num_key_value_heads": 2,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"router_aux_loss_coef": 0.001,
|
||||
"shared_expert_intermediate_size": 512,
|
||||
"use_cache": true,
|
||||
"vocab_size": 248320,
|
||||
"mamba_ssm_dtype": "float32",
|
||||
"rope_parameters": {
|
||||
"mrope_interleaved": true,
|
||||
"mrope_section": [
|
||||
11,
|
||||
11,
|
||||
10
|
||||
],
|
||||
"rope_type": "default",
|
||||
"rope_theta": 10000000,
|
||||
"partial_rotary_factor": 0.25
|
||||
}
|
||||
},
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "4.57.0.dev0",
|
||||
"video_token_id": 248057,
|
||||
"vision_config": {
|
||||
"deepstack_visual_indexes": [],
|
||||
"depth": 27,
|
||||
"hidden_act": "gelu_pytorch_tanh",
|
||||
"hidden_size": 1152,
|
||||
"in_channels": 3,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "qwen3_5_moe",
|
||||
"num_heads": 16,
|
||||
"num_position_embeddings": 2304,
|
||||
"out_hidden_size": 2048,
|
||||
"patch_size": 16,
|
||||
"spatial_merge_size": 2,
|
||||
"temporal_patch_size": 2
|
||||
},
|
||||
"vision_end_token_id": 248054,
|
||||
"vision_start_token_id": 248053,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*visual.*",
|
||||
"re:.*mtp.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*linear_attn.*",
|
||||
"re:model.language_model.layers.\\d+.mlp.gate\\..*",
|
||||
"re:.*shared_expert.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
1813
dev/online_quant_configs/Qwen3.5-35B-A3B-w4a8/quant_map.json
Normal file
1813
dev/online_quant_configs/Qwen3.5-35B-A3B-w4a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
162
dev/online_quant_configs/Qwen3.5-35B-A3B-w8a8/config.json
Normal file
162
dev/online_quant_configs/Qwen3.5-35B-A3B-w8a8/config.json
Normal file
@@ -0,0 +1,162 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3_5MoeForConditionalGeneration"
|
||||
],
|
||||
"image_token_id": 248056,
|
||||
"model_type": "qwen3_5_moe",
|
||||
"text_config": {
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"attn_output_gate": true,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": 248044,
|
||||
"full_attention_interval": 4,
|
||||
"head_dim": 256,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2048,
|
||||
"initializer_range": 0.02,
|
||||
"layer_types": [
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"linear_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"linear_conv_kernel_dim": 4,
|
||||
"linear_key_head_dim": 128,
|
||||
"linear_num_key_heads": 16,
|
||||
"linear_num_value_heads": 32,
|
||||
"linear_value_head_dim": 128,
|
||||
"max_position_embeddings": 262144,
|
||||
"mlp_only_layers": [],
|
||||
"model_type": "qwen3_5_moe_text",
|
||||
"moe_intermediate_size": 512,
|
||||
"mtp_num_hidden_layers": 1,
|
||||
"mtp_use_dedicated_embeddings": false,
|
||||
"num_attention_heads": 16,
|
||||
"num_experts": 256,
|
||||
"num_experts_per_tok": 8,
|
||||
"num_hidden_layers": 40,
|
||||
"num_key_value_heads": 2,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"router_aux_loss_coef": 0.001,
|
||||
"shared_expert_intermediate_size": 512,
|
||||
"use_cache": true,
|
||||
"vocab_size": 248320,
|
||||
"mamba_ssm_dtype": "float32",
|
||||
"rope_parameters": {
|
||||
"mrope_interleaved": true,
|
||||
"mrope_section": [
|
||||
11,
|
||||
11,
|
||||
10
|
||||
],
|
||||
"rope_type": "default",
|
||||
"rope_theta": 10000000,
|
||||
"partial_rotary_factor": 0.25
|
||||
}
|
||||
},
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "4.57.0.dev0",
|
||||
"video_token_id": 248057,
|
||||
"vision_config": {
|
||||
"deepstack_visual_indexes": [],
|
||||
"depth": 27,
|
||||
"hidden_act": "gelu_pytorch_tanh",
|
||||
"hidden_size": 1152,
|
||||
"in_channels": 3,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "qwen3_5_moe",
|
||||
"num_heads": 16,
|
||||
"num_position_embeddings": 2304,
|
||||
"out_hidden_size": 2048,
|
||||
"patch_size": 16,
|
||||
"spatial_merge_size": 2,
|
||||
"temporal_patch_size": 2
|
||||
},
|
||||
"vision_end_token_id": 248054,
|
||||
"vision_start_token_id": 248053,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:lm_head.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*visual.*",
|
||||
"re:.*mtp.*",
|
||||
"re:.*norm.*",
|
||||
"re:.*linear_attn.*",
|
||||
"re:model.language_model.layers.\\d+.mlp.gate\\..*",
|
||||
"re:.*shared_expert.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
1813
dev/online_quant_configs/Qwen3.5-35B-A3B-w8a8/quant_map.json
Normal file
1813
dev/online_quant_configs/Qwen3.5-35B-A3B-w8a8/quant_map.json
Normal file
File diff suppressed because it is too large
Load Diff
362
dev/online_quant_configs/Step-3.5-Flash-w4a8/config.json
Normal file
362
dev/online_quant_configs/Step-3.5-Flash-w4a8/config.json
Normal file
@@ -0,0 +1,362 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Step3p5ForCausalLM"
|
||||
],
|
||||
"model_type": "step3p5",
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_step3p5.Step3p5Config",
|
||||
"AutoModelForCausalLM": "modeling_step3p5.Step3p5ForCausalLM"
|
||||
},
|
||||
"rope_scaling": {
|
||||
"rope_type": "llama3",
|
||||
"factor": 2.0,
|
||||
"original_max_position_embeddings": 131072,
|
||||
"low_freq_factor": 1.0,
|
||||
"high_freq_factor": 32.0
|
||||
},
|
||||
"yarn_only_types": [
|
||||
"full_attention"
|
||||
],
|
||||
"hidden_size": 4096,
|
||||
"intermediate_size": 11264,
|
||||
"num_hidden_layers": 45,
|
||||
"max_seq_len": 262144,
|
||||
"vocab_size": 128896,
|
||||
"torch_dtype": "bfloat16",
|
||||
"use_qk_norm": true,
|
||||
"moe_layers_enum": "3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44",
|
||||
"num_attention_heads": 64,
|
||||
"num_attention_groups": 8,
|
||||
"head_dim": 128,
|
||||
"use_moe": true,
|
||||
"moe_num_experts": 288,
|
||||
"moe_top_k": 8,
|
||||
"moe_intermediate_size": 1280,
|
||||
"share_expert_dim": 1280,
|
||||
"moe_layer_offset": 0,
|
||||
"moe_every_n_layer": 1,
|
||||
"norm_expert_weight": true,
|
||||
"moe_router_activation": "sigmoid",
|
||||
"moe_router_scaling_factor": 3.0,
|
||||
"att_impl_type": "GQA",
|
||||
"tie_word_embeddings": false,
|
||||
"rope_theta": [
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0
|
||||
],
|
||||
"use_head_wise_attn_gate": true,
|
||||
"sliding_window": 512,
|
||||
"use_moe_router_bias": true,
|
||||
"need_fp32_gate": true,
|
||||
"sink": false,
|
||||
"layer_types": [
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention"
|
||||
],
|
||||
"use_rope_layers": [],
|
||||
"num_nextn_predict_layers": 3,
|
||||
"partial_rotary_factors": [
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0
|
||||
],
|
||||
"eos_token_id": [
|
||||
1,
|
||||
2,
|
||||
128007
|
||||
],
|
||||
"bos_token_id": 0,
|
||||
"attention_other_setting": {
|
||||
"attention_type": "sliding_attention",
|
||||
"num_attention_heads": 96,
|
||||
"num_attention_groups": 8,
|
||||
"head_dim": 128,
|
||||
"true_head_dim": 128
|
||||
},
|
||||
"swiglu_limits": [
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
7,
|
||||
7,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0
|
||||
],
|
||||
"swiglu_limits_shared": [
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
16,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0
|
||||
],
|
||||
"zero_centered": true,
|
||||
"max_position_embeddings": 262144,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*norm.*",
|
||||
"re:.*bias.*",
|
||||
"re:.*moe.gate\\..*",
|
||||
"re:.*eh_proj.*",
|
||||
"re:.*transformer.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*share_expert.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
806
dev/online_quant_configs/Step-3.5-Flash-w4a8/quant_map.json
Normal file
806
dev/online_quant_configs/Step-3.5-Flash-w4a8/quant_map.json
Normal file
@@ -0,0 +1,806 @@
|
||||
{
|
||||
"lm_head.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.24.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.24.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.25.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.26.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.27.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.28.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.29.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.30.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.31.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.32.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.33.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.34.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.35.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.36.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.37.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.38.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.39.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.40.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.41.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.42.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.43.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.44.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.45.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.45.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.45.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.45.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.layers.46.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.46.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.46.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.46.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.46.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.46.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.layers.47.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.47.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.47.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.47.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.47.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.47.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.moe.router_bias": "torch.float32",
|
||||
"model.layers.11.moe.router_bias": "torch.float32",
|
||||
"model.layers.12.moe.router_bias": "torch.float32",
|
||||
"model.layers.13.moe.router_bias": "torch.float32",
|
||||
"model.layers.14.moe.router_bias": "torch.float32",
|
||||
"model.layers.15.moe.router_bias": "torch.float32",
|
||||
"model.layers.16.moe.router_bias": "torch.float32",
|
||||
"model.layers.17.moe.router_bias": "torch.float32",
|
||||
"model.layers.18.moe.router_bias": "torch.float32",
|
||||
"model.layers.19.moe.router_bias": "torch.float32",
|
||||
"model.layers.20.moe.router_bias": "torch.float32",
|
||||
"model.layers.21.moe.router_bias": "torch.float32",
|
||||
"model.layers.22.moe.router_bias": "torch.float32",
|
||||
"model.layers.23.moe.router_bias": "torch.float32",
|
||||
"model.layers.24.moe.router_bias": "torch.float32",
|
||||
"model.layers.25.moe.router_bias": "torch.float32",
|
||||
"model.layers.26.moe.router_bias": "torch.float32",
|
||||
"model.layers.27.moe.router_bias": "torch.float32",
|
||||
"model.layers.28.moe.router_bias": "torch.float32",
|
||||
"model.layers.29.moe.router_bias": "torch.float32",
|
||||
"model.layers.3.moe.router_bias": "torch.float32",
|
||||
"model.layers.30.moe.router_bias": "torch.float32",
|
||||
"model.layers.31.moe.router_bias": "torch.float32",
|
||||
"model.layers.32.moe.router_bias": "torch.float32",
|
||||
"model.layers.33.moe.router_bias": "torch.float32",
|
||||
"model.layers.34.moe.router_bias": "torch.float32",
|
||||
"model.layers.35.moe.router_bias": "torch.float32",
|
||||
"model.layers.36.moe.router_bias": "torch.float32",
|
||||
"model.layers.37.moe.router_bias": "torch.float32",
|
||||
"model.layers.38.moe.router_bias": "torch.float32",
|
||||
"model.layers.39.moe.router_bias": "torch.float32",
|
||||
"model.layers.4.moe.router_bias": "torch.float32",
|
||||
"model.layers.40.moe.router_bias": "torch.float32",
|
||||
"model.layers.41.moe.router_bias": "torch.float32",
|
||||
"model.layers.42.moe.router_bias": "torch.float32",
|
||||
"model.layers.43.moe.router_bias": "torch.float32",
|
||||
"model.layers.44.moe.router_bias": "torch.float32",
|
||||
"model.layers.5.moe.router_bias": "torch.float32",
|
||||
"model.layers.6.moe.router_bias": "torch.float32",
|
||||
"model.layers.7.moe.router_bias": "torch.float32",
|
||||
"model.layers.8.moe.router_bias": "torch.float32",
|
||||
"model.layers.9.moe.router_bias": "torch.float32",
|
||||
"model.embed_tokens.weight": "torch.bfloat16",
|
||||
"model.layers.0.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.0.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.0.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.0.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.0.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.1.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.1.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.1.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.1.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.1.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.10.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.10.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.10.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.10.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.10.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.10.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.11.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.11.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.11.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.12.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.12.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.12.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.13.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.13.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.13.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.14.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.14.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.14.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.15.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.15.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.15.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.16.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.16.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.16.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.17.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.17.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.17.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.18.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.18.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.18.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.19.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.19.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.19.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.2.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.2.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.2.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.2.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.2.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.20.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.20.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.20.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.20.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.20.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.20.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.21.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.21.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.21.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.22.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.22.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.22.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.23.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.23.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.23.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.24.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.24.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.24.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.24.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.25.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.25.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.25.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.26.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.26.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.26.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.27.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.27.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.27.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.28.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.28.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.28.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.29.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.29.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.29.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.3.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.3.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.3.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.30.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.30.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.30.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.31.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.31.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.31.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.32.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.32.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.32.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.33.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.33.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.33.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.34.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.34.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.34.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.35.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.35.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.35.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.36.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.36.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.36.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.37.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.37.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.37.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.38.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.38.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.38.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.39.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.39.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.39.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.4.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.4.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.4.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.40.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.40.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.40.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.41.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.41.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.41.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.42.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.42.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.42.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.43.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.43.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.43.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.44.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.44.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.44.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.5.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.5.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.5.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.5.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.5.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.6.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.6.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.6.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.7.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.7.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.7.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.8.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.8.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.8.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.9.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.9.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.9.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.11.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.11.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.12.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.12.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.12.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.10.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.10.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.10.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.5.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.5.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.5.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.3.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.3.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.3.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.14.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.14.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.14.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.16.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.16.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.16.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.7.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.7.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.7.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.8.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.8.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.8.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.4.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.4.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.4.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.15.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.15.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.15.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.13.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.13.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.13.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.6.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.6.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.6.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.9.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.9.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.9.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.17.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.17.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.17.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.18.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.18.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.18.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.19.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.19.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.19.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.20.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.20.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.20.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.21.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.21.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.21.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.22.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.22.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.22.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.23.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.23.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.23.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.24.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.24.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.24.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.25.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.25.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.25.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.26.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.26.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.26.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.27.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.27.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.27.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.28.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.28.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.28.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.29.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.29.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.29.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.30.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.30.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.30.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.31.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.31.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.31.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.32.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.32.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.32.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.33.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.33.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.33.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.34.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.34.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.34.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.35.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.35.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.35.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.36.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.36.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.36.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.37.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.37.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.37.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.38.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.38.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.38.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.39.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.39.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.39.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.40.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.40.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.40.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.41.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.41.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.41.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.42.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.42.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.42.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.43.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.43.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.43.moe.up_proj.weight": "int4-TN",
|
||||
"model.layers.44.moe.down_proj.weight": "int4-TN",
|
||||
"model.layers.44.moe.gate_proj.weight": "int4-TN",
|
||||
"model.layers.44.moe.up_proj.weight": "int4-TN"
|
||||
}
|
||||
362
dev/online_quant_configs/Step-3.5-Flash-w8a8/config.json
Normal file
362
dev/online_quant_configs/Step-3.5-Flash-w8a8/config.json
Normal file
@@ -0,0 +1,362 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Step3p5ForCausalLM"
|
||||
],
|
||||
"model_type": "step3p5",
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_step3p5.Step3p5Config",
|
||||
"AutoModelForCausalLM": "modeling_step3p5.Step3p5ForCausalLM"
|
||||
},
|
||||
"rope_scaling": {
|
||||
"rope_type": "llama3",
|
||||
"factor": 2.0,
|
||||
"original_max_position_embeddings": 131072,
|
||||
"low_freq_factor": 1.0,
|
||||
"high_freq_factor": 32.0
|
||||
},
|
||||
"yarn_only_types": [
|
||||
"full_attention"
|
||||
],
|
||||
"hidden_size": 4096,
|
||||
"intermediate_size": 11264,
|
||||
"num_hidden_layers": 45,
|
||||
"max_seq_len": 262144,
|
||||
"vocab_size": 128896,
|
||||
"torch_dtype": "bfloat16",
|
||||
"use_qk_norm": true,
|
||||
"moe_layers_enum": "3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44",
|
||||
"num_attention_heads": 64,
|
||||
"num_attention_groups": 8,
|
||||
"head_dim": 128,
|
||||
"use_moe": true,
|
||||
"moe_num_experts": 288,
|
||||
"moe_top_k": 8,
|
||||
"moe_intermediate_size": 1280,
|
||||
"share_expert_dim": 1280,
|
||||
"moe_layer_offset": 0,
|
||||
"moe_every_n_layer": 1,
|
||||
"norm_expert_weight": true,
|
||||
"moe_router_activation": "sigmoid",
|
||||
"moe_router_scaling_factor": 3.0,
|
||||
"att_impl_type": "GQA",
|
||||
"tie_word_embeddings": false,
|
||||
"rope_theta": [
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
5000000.0,
|
||||
10000.0,
|
||||
10000.0,
|
||||
10000.0
|
||||
],
|
||||
"use_head_wise_attn_gate": true,
|
||||
"sliding_window": 512,
|
||||
"use_moe_router_bias": true,
|
||||
"need_fp32_gate": true,
|
||||
"sink": false,
|
||||
"layer_types": [
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"full_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention",
|
||||
"sliding_attention"
|
||||
],
|
||||
"use_rope_layers": [],
|
||||
"num_nextn_predict_layers": 3,
|
||||
"partial_rotary_factors": [
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0,
|
||||
0.5,
|
||||
1.0,
|
||||
1.0,
|
||||
1.0
|
||||
],
|
||||
"eos_token_id": [
|
||||
1,
|
||||
2,
|
||||
128007
|
||||
],
|
||||
"bos_token_id": 0,
|
||||
"attention_other_setting": {
|
||||
"attention_type": "sliding_attention",
|
||||
"num_attention_heads": 96,
|
||||
"num_attention_groups": 8,
|
||||
"head_dim": 128,
|
||||
"true_head_dim": 128
|
||||
},
|
||||
"swiglu_limits": [
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
7,
|
||||
7,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0
|
||||
],
|
||||
"swiglu_limits_shared": [
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0,
|
||||
16,
|
||||
0.0,
|
||||
0.0,
|
||||
0.0
|
||||
],
|
||||
"zero_centered": true,
|
||||
"max_position_embeddings": 262144,
|
||||
"quantization_config": {
|
||||
"config_groups": {
|
||||
"group_0": {
|
||||
"input_activations": {
|
||||
"block_structure": null,
|
||||
"dynamic": true,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "memoryless",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "token",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
},
|
||||
"output_activations": null,
|
||||
"targets": [
|
||||
"Linear"
|
||||
],
|
||||
"weights": {
|
||||
"block_structure": null,
|
||||
"dynamic": false,
|
||||
"group_size": null,
|
||||
"num_bits": 8,
|
||||
"observer": "minmax",
|
||||
"observer_kwargs": {},
|
||||
"strategy": "channel",
|
||||
"symmetric": true,
|
||||
"type": "int"
|
||||
}
|
||||
}
|
||||
},
|
||||
"format": "int-quantized",
|
||||
"global_compression_ratio": 1.0,
|
||||
"ignore": [
|
||||
"re:.*norm.*",
|
||||
"re:.*bias.*",
|
||||
"re:.*moe.gate\\..*",
|
||||
"re:.*eh_proj.*",
|
||||
"re:.*transformer.*",
|
||||
"re:.*lm_head.*",
|
||||
"re:.*embed_tokens.*",
|
||||
"re:.*share_expert.*"
|
||||
],
|
||||
"kv_cache_scheme": null,
|
||||
"quant_method": "compressed-tensors",
|
||||
"quantization_status": "frozen"
|
||||
}
|
||||
}
|
||||
806
dev/online_quant_configs/Step-3.5-Flash-w8a8/quant_map.json
Normal file
806
dev/online_quant_configs/Step-3.5-Flash-w8a8/quant_map.json
Normal file
@@ -0,0 +1,806 @@
|
||||
{
|
||||
"lm_head.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.24.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.24.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.25.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.26.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.27.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.28.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.29.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.30.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.31.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.32.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.33.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.34.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.35.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.36.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.37.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.38.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.39.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.40.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.41.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.42.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.43.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.44.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.45.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.45.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.45.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.45.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.45.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.45.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.layers.46.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.46.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.46.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.46.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.46.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.46.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.46.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.46.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.layers.47.eh_proj.weight": "torch.bfloat16",
|
||||
"model.layers.47.enorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.hnorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.47.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.47.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.47.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.47.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.47.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.47.transformer.shared_head.output.weight": "torch.bfloat16",
|
||||
"model.norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.moe.router_bias": "torch.float32",
|
||||
"model.layers.11.moe.router_bias": "torch.float32",
|
||||
"model.layers.12.moe.router_bias": "torch.float32",
|
||||
"model.layers.13.moe.router_bias": "torch.float32",
|
||||
"model.layers.14.moe.router_bias": "torch.float32",
|
||||
"model.layers.15.moe.router_bias": "torch.float32",
|
||||
"model.layers.16.moe.router_bias": "torch.float32",
|
||||
"model.layers.17.moe.router_bias": "torch.float32",
|
||||
"model.layers.18.moe.router_bias": "torch.float32",
|
||||
"model.layers.19.moe.router_bias": "torch.float32",
|
||||
"model.layers.20.moe.router_bias": "torch.float32",
|
||||
"model.layers.21.moe.router_bias": "torch.float32",
|
||||
"model.layers.22.moe.router_bias": "torch.float32",
|
||||
"model.layers.23.moe.router_bias": "torch.float32",
|
||||
"model.layers.24.moe.router_bias": "torch.float32",
|
||||
"model.layers.25.moe.router_bias": "torch.float32",
|
||||
"model.layers.26.moe.router_bias": "torch.float32",
|
||||
"model.layers.27.moe.router_bias": "torch.float32",
|
||||
"model.layers.28.moe.router_bias": "torch.float32",
|
||||
"model.layers.29.moe.router_bias": "torch.float32",
|
||||
"model.layers.3.moe.router_bias": "torch.float32",
|
||||
"model.layers.30.moe.router_bias": "torch.float32",
|
||||
"model.layers.31.moe.router_bias": "torch.float32",
|
||||
"model.layers.32.moe.router_bias": "torch.float32",
|
||||
"model.layers.33.moe.router_bias": "torch.float32",
|
||||
"model.layers.34.moe.router_bias": "torch.float32",
|
||||
"model.layers.35.moe.router_bias": "torch.float32",
|
||||
"model.layers.36.moe.router_bias": "torch.float32",
|
||||
"model.layers.37.moe.router_bias": "torch.float32",
|
||||
"model.layers.38.moe.router_bias": "torch.float32",
|
||||
"model.layers.39.moe.router_bias": "torch.float32",
|
||||
"model.layers.4.moe.router_bias": "torch.float32",
|
||||
"model.layers.40.moe.router_bias": "torch.float32",
|
||||
"model.layers.41.moe.router_bias": "torch.float32",
|
||||
"model.layers.42.moe.router_bias": "torch.float32",
|
||||
"model.layers.43.moe.router_bias": "torch.float32",
|
||||
"model.layers.44.moe.router_bias": "torch.float32",
|
||||
"model.layers.5.moe.router_bias": "torch.float32",
|
||||
"model.layers.6.moe.router_bias": "torch.float32",
|
||||
"model.layers.7.moe.router_bias": "torch.float32",
|
||||
"model.layers.8.moe.router_bias": "torch.float32",
|
||||
"model.layers.9.moe.router_bias": "torch.float32",
|
||||
"model.embed_tokens.weight": "torch.bfloat16",
|
||||
"model.layers.0.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.0.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.0.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.0.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.0.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.0.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.0.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.1.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.1.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.1.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.1.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.1.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.1.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.1.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.10.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.10.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.10.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.10.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.10.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.10.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.10.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.10.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.11.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.11.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.11.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.11.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.11.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.11.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.12.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.12.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.12.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.12.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.12.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.12.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.13.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.13.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.13.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.13.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.13.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.13.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.14.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.14.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.14.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.14.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.14.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.14.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.15.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.15.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.15.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.15.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.15.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.15.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.16.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.16.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.16.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.16.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.16.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.16.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.17.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.17.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.17.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.17.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.17.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.17.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.18.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.18.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.18.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.18.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.18.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.18.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.19.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.19.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.19.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.19.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.19.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.19.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.2.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.2.mlp.down_proj.weight": "int8-TN",
|
||||
"model.layers.2.mlp.gate_proj.weight": "int8-TN",
|
||||
"model.layers.2.mlp.up_proj.weight": "int8-TN",
|
||||
"model.layers.2.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.2.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.2.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.20.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.20.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.20.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.20.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.20.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.20.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.20.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.20.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.21.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.21.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.21.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.21.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.21.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.21.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.22.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.22.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.22.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.22.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.22.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.22.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.23.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.23.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.23.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.23.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.23.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.23.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.24.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.24.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.24.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.24.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.24.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.24.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.25.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.25.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.25.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.25.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.25.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.25.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.26.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.26.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.26.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.26.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.26.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.26.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.27.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.27.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.27.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.27.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.27.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.27.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.28.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.28.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.28.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.28.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.28.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.28.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.29.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.29.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.29.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.29.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.29.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.29.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.3.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.3.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.3.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.3.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.3.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.3.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.30.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.30.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.30.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.30.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.30.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.30.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.31.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.31.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.31.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.31.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.31.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.31.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.32.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.32.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.32.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.32.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.32.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.32.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.33.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.33.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.33.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.33.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.33.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.33.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.34.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.34.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.34.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.34.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.34.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.34.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.35.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.35.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.35.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.35.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.35.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.35.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.36.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.36.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.36.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.36.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.36.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.36.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.37.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.37.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.37.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.37.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.37.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.37.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.38.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.38.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.38.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.38.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.38.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.38.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.39.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.39.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.39.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.39.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.39.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.39.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.4.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.4.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.4.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.4.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.4.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.4.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.40.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.40.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.40.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.40.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.40.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.40.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.41.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.41.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.41.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.41.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.41.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.41.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.42.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.42.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.42.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.42.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.42.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.42.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.43.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.43.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.43.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.43.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.43.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.43.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.44.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.44.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.44.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.44.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.44.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.44.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.45.transformer.shared_head.norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.5.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.5.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.5.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.5.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.5.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.5.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.5.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.6.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.6.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.6.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.6.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.6.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.6.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.7.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.7.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.7.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.7.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.7.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.7.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.8.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.8.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.8.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.8.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.8.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.8.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.input_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.9.moe.gate.weight": "torch.bfloat16",
|
||||
"model.layers.9.post_attention_layernorm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.g_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.k_norm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.k_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.o_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.q_norm.weight": "torch.bfloat16",
|
||||
"model.layers.9.self_attn.q_proj.weight": "int8-TN",
|
||||
"model.layers.9.self_attn.v_proj.weight": "int8-TN",
|
||||
"model.layers.9.share_expert.down_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.share_expert.gate_proj.weight": "torch.bfloat16",
|
||||
"model.layers.9.share_expert.up_proj.weight": "torch.bfloat16",
|
||||
"model.layers.5.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.5.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.5.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.10.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.10.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.10.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.14.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.14.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.14.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.4.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.4.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.4.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.6.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.6.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.6.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.7.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.7.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.7.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.11.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.11.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.11.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.13.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.13.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.13.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.12.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.12.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.12.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.9.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.9.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.9.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.8.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.8.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.8.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.16.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.16.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.16.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.3.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.3.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.3.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.15.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.15.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.15.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.17.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.17.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.17.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.18.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.18.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.18.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.19.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.19.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.19.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.20.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.20.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.20.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.21.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.21.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.21.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.22.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.22.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.22.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.23.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.23.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.23.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.24.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.24.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.24.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.25.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.25.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.25.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.26.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.26.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.26.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.27.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.27.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.27.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.28.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.28.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.28.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.29.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.29.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.29.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.30.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.30.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.30.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.31.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.31.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.31.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.32.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.32.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.32.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.33.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.33.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.33.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.34.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.34.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.34.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.35.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.35.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.35.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.36.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.36.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.36.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.37.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.37.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.37.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.38.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.38.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.38.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.39.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.39.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.39.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.40.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.40.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.40.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.41.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.41.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.41.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.42.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.42.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.42.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.43.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.43.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.43.moe.up_proj.weight": "int8-TN",
|
||||
"model.layers.44.moe.down_proj.weight": "int8-TN",
|
||||
"model.layers.44.moe.gate_proj.weight": "int8-TN",
|
||||
"model.layers.44.moe.up_proj.weight": "int8-TN"
|
||||
}
|
||||
Reference in New Issue
Block a user