Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,82 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
---
|
| 4 |
+
# 0506_7_7
|
| 5 |
+
|
| 6 |
+
This model is a fine-tuned version of [../../models/Qwen1.5-7B-sft-0502](https://huggingface.co/../../models/Qwen1.5-7B-sft-0502) on the alpaca_formatted_review_new_data_0505_greater_7 dataset.
|
| 7 |
+
It achieves the following results on the evaluation set:
|
| 8 |
+
|
| 9 |
+
- Loss: 0.7221
|
| 10 |
+
|
| 11 |
+
## Model description
|
| 12 |
+
|
| 13 |
+
Qwen1.5 is the beta version of Qwen2, a transformer-based decoder-only language model pretrained on a large amount of data. In comparison with the previous released Qwen, the improvements include:
|
| 14 |
+
|
| 15 |
+
* 8 model sizes, including 0.5B, 1.8B, 4B, 7B, 14B, 32B and 72B dense models, and an MoE model of 14B with 2.7B activated;
|
| 16 |
+
* Significant performance improvement in Chat models;
|
| 17 |
+
* Multilingual support of both base and chat models;
|
| 18 |
+
* Stable support of 32K context length for models of all sizes
|
| 19 |
+
* No need of `trust_remote_code`.
|
| 20 |
+
|
| 21 |
+
For more details, please refer to the [blog post](https://qwenlm.github.io/blog/qwen1.5/) and [GitHub repo](https://github.com/QwenLM/Qwen1.5).
|
| 22 |
+
|
| 23 |
+
## Intended uses & limitations
|
| 24 |
+
|
| 25 |
+
More information needed
|
| 26 |
+
|
| 27 |
+
## Training and evaluation data
|
| 28 |
+
|
| 29 |
+
More information needed
|
| 30 |
+
|
| 31 |
+
## Training procedure
|
| 32 |
+
|
| 33 |
+
### Training hyperparameters
|
| 34 |
+
|
| 35 |
+
The following hyperparameters were used during training:
|
| 36 |
+
|
| 37 |
+
- learning_rate: 0.0003
|
| 38 |
+
- train_batch_size: 2
|
| 39 |
+
- eval_batch_size: 1
|
| 40 |
+
- seed: 42
|
| 41 |
+
- distributed_type: multi-GPU
|
| 42 |
+
- num_devices: 2
|
| 43 |
+
- gradient_accumulation_steps: 4
|
| 44 |
+
- total_train_batch_size: 16
|
| 45 |
+
- total_eval_batch_size: 2
|
| 46 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 47 |
+
- lr_scheduler_type: cosine
|
| 48 |
+
- lr_scheduler_warmup_steps: 20
|
| 49 |
+
- num_epochs: 5.0
|
| 50 |
+
- mixed_precision_training: Native AMP
|
| 51 |
+
|
| 52 |
+
### Training results
|
| 53 |
+
|
| 54 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
| 55 |
+
| :-----------: | :----: | :--: | :-------------: |
|
| 56 |
+
| 0.7981 | 0.2768 | 20 | 0.6501 |
|
| 57 |
+
| 0.7391 | 0.5536 | 40 | 0.6358 |
|
| 58 |
+
| 0.744 | 0.8304 | 60 | 0.6277 |
|
| 59 |
+
| 0.6284 | 1.1073 | 80 | 0.6241 |
|
| 60 |
+
| 0.7339 | 1.3841 | 100 | 0.6303 |
|
| 61 |
+
| 0.8346 | 1.6609 | 120 | 0.6408 |
|
| 62 |
+
| 0.6927 | 1.9377 | 140 | 0.6391 |
|
| 63 |
+
| 0.4915 | 2.2145 | 160 | 0.6543 |
|
| 64 |
+
| 0.7845 | 2.4913 | 180 | 0.6596 |
|
| 65 |
+
| 0.6619 | 2.7682 | 200 | 0.6587 |
|
| 66 |
+
| 0.4897 | 3.0450 | 220 | 0.6679 |
|
| 67 |
+
| 0.5064 | 3.3218 | 240 | 0.6951 |
|
| 68 |
+
| 0.6467 | 3.5986 | 260 | 0.6997 |
|
| 69 |
+
| 0.6615 | 3.8754 | 280 | 0.6985 |
|
| 70 |
+
| 0.4954 | 4.1522 | 300 | 0.7111 |
|
| 71 |
+
| 0.5624 | 4.4291 | 320 | 0.7216 |
|
| 72 |
+
| 0.5554 | 4.7059 | 340 | 0.7218 |
|
| 73 |
+
| 0.6798 | 4.9827 | 360 | 0.7221 |
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
### Framework versions
|
| 77 |
+
|
| 78 |
+
- PEFT 0.10.0
|
| 79 |
+
- Transformers 4.40.0
|
| 80 |
+
- Pytorch 2.1.0+cu121
|
| 81 |
+
- Datasets 2.14.5
|
| 82 |
+
- Tokenizers 0.19.1
|