128 lines
3.4 KiB
YAML
128 lines
3.4 KiB
YAML
|
base_model: huggyllama/llama-30b
|
||
|
|
base_model_config: huggyllama/llama-30b
|
||
|
|
model_type: LlamaForCausalLM
|
||
|
|
tokenizer_type: LlamaTokenizer
|
||
|
|
load_in_8bit: false
|
||
|
|
load_in_4bit: false
|
||
|
|
strict: false
|
||
|
|
push_dataset_to_hub: winglian
|
||
|
|
hf_use_auth_token: true
|
||
|
|
datasets:
|
||
|
|
- path: winglian/pygmalion-cleaned
|
||
|
|
data_files:
|
||
|
|
- v13_no_ai.cleaned.jsonl
|
||
|
|
type: pygmalion
|
||
|
|
- path: winglian/evals
|
||
|
|
data_files:
|
||
|
|
- hf/ARC-Challenge.jsonl
|
||
|
|
- hf/ARC-Easy.jsonl
|
||
|
|
- hf/riddle_sense.jsonl
|
||
|
|
type: explainchoice:chat
|
||
|
|
- path: winglian/evals
|
||
|
|
data_files:
|
||
|
|
- hf/gsm8k.jsonl
|
||
|
|
- custom/logic_inference_oa.jsonl
|
||
|
|
type: alpaca_chat.load_qa
|
||
|
|
- 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
|
||
|
|
- path: winglian/evals
|
||
|
|
data_files:
|
||
|
|
- custom/guanaco-cleaned.en.jsonl
|
||
|
|
type: sharegpt_simple.load_guanaco
|
||
|
|
- path: winglian/evals
|
||
|
|
data_files:
|
||
|
|
- openai/tldr.jsonl
|
||
|
|
type: summarizetldr:chat
|
||
|
|
- path: winglian/evals
|
||
|
|
data_files:
|
||
|
|
- hellaswag/hellaswag.jsonl
|
||
|
|
type: explainchoice:chat
|
||
|
|
shards: 60
|
||
|
|
- path: metaeval/ScienceQA_text_only
|
||
|
|
type: concisechoice:chat
|
||
|
|
shards: 13
|
||
|
|
- 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/gsm8k_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
|
||
|
|
dataset_prepared_path: /workspace/mnt/last_run_prepared
|
||
|
|
val_set_size: 0.02
|
||
|
|
adapter:
|
||
|
|
lora_model_dir:
|
||
|
|
sequence_len: 2048
|
||
|
|
max_packed_sequence_len: 2048
|
||
|
|
lora_r: 64
|
||
|
|
lora_alpha: 32
|
||
|
|
lora_dropout: 0.000001
|
||
|
|
lora_target_modules:
|
||
|
|
lora_target_linear: true
|
||
|
|
lora_fan_in_fan_out:
|
||
|
|
wandb_project: hippogriff-30b-chat
|
||
|
|
wandb_watch:
|
||
|
|
wandb_run_id:
|
||
|
|
wandb_log_model:
|
||
|
|
output_dir: /workspace/mnt/hippogriff-30b-chat
|
||
|
|
gradient_accumulation_steps: 1
|
||
|
|
micro_batch_size: 20
|
||
|
|
num_epochs: 2
|
||
|
|
optimizer:
|
||
|
|
torchdistx_path:
|
||
|
|
lr_scheduler:
|
||
|
|
learning_rate: 0.00003
|
||
|
|
train_on_inputs: false
|
||
|
|
group_by_length: true
|
||
|
|
bf16: true
|
||
|
|
fp16: false
|
||
|
|
float16: 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: 200
|
||
|
|
eval_steps: 20
|
||
|
|
save_steps: 50
|
||
|
|
debug:
|
||
|
|
deepspeed:
|
||
|
|
weight_decay: 0.000003
|
||
|
|
fsdp:
|
||
|
|
fsdp_config:
|
||
|
|
special_tokens:
|
||
|
|
bos_token: "<s>"
|
||
|
|
eos_token: "</s>"
|
||
|
|
unk_token: "<unk>"
|