Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- Qwen/Qwen2.5-7B
|
| 5 |
+
library_name: transformers
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
```
|
| 9 |
+
base_model: Qwen/Qwen2.5-7B
|
| 10 |
+
model_type: AutoModelForCausalLM
|
| 11 |
+
tokenizer_type: AutoTokenizer
|
| 12 |
+
|
| 13 |
+
load_in_8bit: false
|
| 14 |
+
load_in_4bit: false
|
| 15 |
+
strict: false
|
| 16 |
+
|
| 17 |
+
datasets:
|
| 18 |
+
- path: PocketDoc/Dans-MemoryCore-CoreCurriculum-Small
|
| 19 |
+
type: sharegpt
|
| 20 |
+
conversation: chatml
|
| 21 |
+
- path: NewEden/Kalo-Opus-Instruct-22k-Refusal-Murdered
|
| 22 |
+
type: sharegpt
|
| 23 |
+
conversation: chatml
|
| 24 |
+
- path: Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
|
| 25 |
+
type: sharegpt
|
| 26 |
+
conversation: chatml
|
| 27 |
+
- path: NewEden/Gryphe-Sonnet-3.5-35k-Subset
|
| 28 |
+
type: sharegpt
|
| 29 |
+
conversation: chatml
|
| 30 |
+
- path: Nitral-AI/Reasoning-1shot_ShareGPT
|
| 31 |
+
type: sharegpt
|
| 32 |
+
conversation: chatml
|
| 33 |
+
- path: Nitral-AI/GU_Instruct-ShareGPT
|
| 34 |
+
type: sharegpt
|
| 35 |
+
conversation: chatml
|
| 36 |
+
- path: Nitral-AI/Medical_Instruct-ShareGPT
|
| 37 |
+
type: sharegpt
|
| 38 |
+
conversation: chatml
|
| 39 |
+
|
| 40 |
+
chat_template: chatml
|
| 41 |
+
|
| 42 |
+
val_set_size: 0.01
|
| 43 |
+
output_dir: ./outputs/out
|
| 44 |
+
|
| 45 |
+
adapter:
|
| 46 |
+
lora_r:
|
| 47 |
+
lora_alpha:
|
| 48 |
+
lora_dropout:
|
| 49 |
+
lora_target_linear:
|
| 50 |
+
|
| 51 |
+
sequence_len: 8192
|
| 52 |
+
# sequence_len: 32768
|
| 53 |
+
sample_packing: true
|
| 54 |
+
eval_sample_packing: false
|
| 55 |
+
pad_to_sequence_len: true
|
| 56 |
+
|
| 57 |
+
plugins:
|
| 58 |
+
- axolotl.integrations.liger.LigerPlugin
|
| 59 |
+
liger_rope: true
|
| 60 |
+
liger_rms_norm: true
|
| 61 |
+
liger_swiglu: true
|
| 62 |
+
liger_fused_linear_cross_entropy: true
|
| 63 |
+
|
| 64 |
+
wandb_project: qwen7B
|
| 65 |
+
wandb_entity:
|
| 66 |
+
wandb_watch:
|
| 67 |
+
wandb_name: qwen7B
|
| 68 |
+
wandb_log_model:
|
| 69 |
+
|
| 70 |
+
gradient_accumulation_steps: 32
|
| 71 |
+
micro_batch_size: 1
|
| 72 |
+
num_epochs: 2
|
| 73 |
+
optimizer: adamw_bnb_8bit
|
| 74 |
+
lr_scheduler: cosine
|
| 75 |
+
learning_rate: 0.00001
|
| 76 |
+
weight_decay: 0.05
|
| 77 |
+
|
| 78 |
+
train_on_inputs: false
|
| 79 |
+
group_by_length: false
|
| 80 |
+
bf16: auto
|
| 81 |
+
fp16:
|
| 82 |
+
tf32: true
|
| 83 |
+
|
| 84 |
+
gradient_checkpointing: true
|
| 85 |
+
early_stopping_patience:
|
| 86 |
+
resume_from_checkpoint:
|
| 87 |
+
local_rank:
|
| 88 |
+
logging_steps: 1
|
| 89 |
+
xformers_attention:
|
| 90 |
+
flash_attention: true
|
| 91 |
+
|
| 92 |
+
warmup_ratio: 0.1
|
| 93 |
+
evals_per_epoch: 4
|
| 94 |
+
eval_table_size:
|
| 95 |
+
eval_max_new_tokens: 128
|
| 96 |
+
saves_per_epoch: 2
|
| 97 |
+
|
| 98 |
+
debug:
|
| 99 |
+
deepspeed:
|
| 100 |
+
fsdp:
|
| 101 |
+
fsdp_config:
|
| 102 |
+
|
| 103 |
+
special_tokens:
|
| 104 |
+
pad_token: <pad>
|
| 105 |
+
```
|