初始化项目,由ModelHub XC社区提供模型
Model: openaccess-ai-collective/minotaur-13b-fixed Source: Original Platform
This commit is contained in:
143
configs/minotaur-13b.yml
Normal file
143
configs/minotaur-13b.yml
Normal file
@@ -0,0 +1,143 @@
|
||||
base_model: huggyllama/llama-13b
|
||||
base_model_config: huggyllama/llama-13b
|
||||
model_type: LlamaForCausalLM
|
||||
tokenizer_type: LlamaTokenizer
|
||||
load_in_8bit: false
|
||||
load_in_4bit: false
|
||||
gptq: false
|
||||
strict: false
|
||||
push_dataset_to_hub: winglian
|
||||
hf_use_auth_token: true
|
||||
datasets:
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- hf/ARC-Challenge.jsonl
|
||||
- hf/ARC-Easy.jsonl
|
||||
- hf/riddle_sense.jsonl
|
||||
- hf/piqa.jsonl
|
||||
type: explainchoice:chat
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- hf/gsm8k.jsonl
|
||||
- hf/winogrande.jsonl
|
||||
type: alpaca_chat.load_qa
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- custom/n_task.jsonl
|
||||
- custom/misconceptions.jsonl
|
||||
- custom/context_insensitivity.jsonl
|
||||
type: alpaca_chat
|
||||
- path: camel-ai/math
|
||||
type: alpaca_chat.load_camel_ai
|
||||
- path: camel-ai/biology
|
||||
type: alpaca_chat.load_camel_ai
|
||||
- path: camel-ai/physics
|
||||
type: alpaca_chat.load_camel_ai
|
||||
- path: camel-ai/chemistry
|
||||
type: alpaca_chat.load_camel_ai
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- custom/in_context_qa.jsonl
|
||||
type: context_qa
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- custom/in_context_qa.jsonl
|
||||
type: context_qa.load_404
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- custom/jokes_explained_500up.jsonl
|
||||
type: sharegpt_jokes
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- custom/classify-self-chat.sharegpt.jsonl
|
||||
- custom/coding-self-chat.sharegpt.jsonl
|
||||
- custom/prose-gpt4.sharegpt.jsonl
|
||||
- custom/prose-rewrite-gpt4.sharegpt.jsonl
|
||||
type: sharegpt_simple.load_role
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- openai/tldr.jsonl
|
||||
type: summarizetldr:chat
|
||||
- path: winglian/evals
|
||||
data_files:
|
||||
- hellaswag/hellaswag.jsonl
|
||||
type: explainchoice:chat
|
||||
- path: metaeval/ScienceQA_text_only
|
||||
type: concisechoice:chat
|
||||
- path: teknium/GPT4-LLM-Cleaned
|
||||
type: alpaca_chat
|
||||
- path: teknium/GPTeacher-General-Instruct
|
||||
data_files: gpt4-instruct-similarity-0.6-dataset.json
|
||||
type: gpteacher:chat
|
||||
- path: QingyiSi/Alpaca-CoT
|
||||
data_files:
|
||||
- Chain-of-Thought/formatted_cot_data/aqua_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/creak_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/ecqa_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/esnli_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/qasc_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/qed_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/sensemaking_train.json
|
||||
- Chain-of-Thought/formatted_cot_data/strategyqa_train.json
|
||||
- GPTeacher/Roleplay/formatted_roleplay-similarity_0.6-instruct-dataset.json
|
||||
type: alpaca_chat
|
||||
- path: ehartford/WizardLM_alpaca_evol_instruct_70k_unfiltered
|
||||
type: alpaca_chat
|
||||
- path: ehartford/wizard_vicuna_70k_unfiltered
|
||||
type: sharegpt:chat
|
||||
dataset_prepared_path: last_run_prepared
|
||||
val_set_size: 0.02
|
||||
adapter:
|
||||
lora_model_dir:
|
||||
sequence_len: 2048
|
||||
max_packed_sequence_len: 2048
|
||||
lora_r:
|
||||
lora_alpha:
|
||||
lora_dropout:
|
||||
lora_target_modules:
|
||||
lora_target_linear: true
|
||||
lora_fan_in_fan_out:
|
||||
wandb_project: minotaur-13b
|
||||
wandb_watch:
|
||||
wandb_run_id:
|
||||
wandb_log_model:
|
||||
output_dir: ./minotaur-13b
|
||||
gradient_accumulation_steps: 1
|
||||
micro_batch_size: 12
|
||||
num_epochs: 3
|
||||
optimizer: adamw_bnb_8bit
|
||||
torchdistx_path:
|
||||
lr_scheduler: cosine
|
||||
learning_rate: 0.00003
|
||||
train_on_inputs: false
|
||||
group_by_length: true
|
||||
bf16: true
|
||||
fp16: false
|
||||
tf32: true
|
||||
gradient_checkpointing: true
|
||||
early_stopping_patience:
|
||||
resume_from_checkpoint:
|
||||
local_rank:
|
||||
logging_steps: 1
|
||||
xformers_attention: true
|
||||
flash_attention:
|
||||
gptq_groupsize:
|
||||
gptq_model_v1:
|
||||
warmup_steps: 20
|
||||
eval_steps: 38
|
||||
save_steps: 56
|
||||
load_best_model_at_end: false
|
||||
debug:
|
||||
deepspeed:
|
||||
weight_decay: 0.001
|
||||
fsdp:
|
||||
- full_shard
|
||||
- auto_wrap
|
||||
fsdp_config:
|
||||
fsdp_offload_params: true
|
||||
fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
|
||||
special_tokens:
|
||||
bos_token: "<s>"
|
||||
eos_token: "</s>"
|
||||
unk_token: "<unk>"
|
||||
|
||||
Reference in New Issue
Block a user