Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
This model is a fine-tuned version of Mixtral-8x22B-Instruct-v0.1 on the mbpp dataset.
|
| 2 |
+
|
| 3 |
+
# Model description
|
| 4 |
+
More information needed
|
| 5 |
+
|
| 6 |
+
# Intended uses & limitations
|
| 7 |
+
More information needed
|
| 8 |
+
|
| 9 |
+
# Training and evaluation data
|
| 10 |
+
More information needed
|
| 11 |
+
|
| 12 |
+
# Training hyperparameters
|
| 13 |
+
The following hyperparameters were used during training:
|
| 14 |
+
|
| 15 |
+
### method
|
| 16 |
+
- stage: sft
|
| 17 |
+
- finetuning_type: lora
|
| 18 |
+
- lora_target: all
|
| 19 |
+
- deepspeed: examples/deepspeed/ds_z3_offload_config.json
|
| 20 |
+
|
| 21 |
+
### dataset
|
| 22 |
+
- dataset: mbpp
|
| 23 |
+
- template: mistral
|
| 24 |
+
- cutoff_len: 2048
|
| 25 |
+
- max_samples: 316
|
| 26 |
+
- overwrite_cache: true
|
| 27 |
+
- preprocessing_num_workers: 16
|
| 28 |
+
|
| 29 |
+
### train
|
| 30 |
+
- per_device_train_batch_size: 1
|
| 31 |
+
- gradient_accumulation_steps: 2
|
| 32 |
+
- learning_rate: 1.0e-4
|
| 33 |
+
- num_train_epochs: 3
|
| 34 |
+
- lr_scheduler_type: cosine
|
| 35 |
+
- warmup_ratio: 0.1
|
| 36 |
+
- bf16: true
|
| 37 |
+
- ddp_timeout: 180000000
|
| 38 |
+
|
| 39 |
+
# Framework versions
|
| 40 |
+
- PEFT 0.14.0
|
| 41 |
+
- Transformers 4.47.0
|
| 42 |
+
- Pytorch 2.5.1+cu124
|
| 43 |
+
- Datasets 2.14.6
|
| 44 |
+
- Tokenizers 0.21.0
|