Improve language tag
#1
by
lbourdois
- opened
README.md
CHANGED
|
@@ -1,46 +1,60 @@
|
|
| 1 |
-
---
|
| 2 |
-
library_name: transformers
|
| 3 |
-
license: apache-2.0
|
| 4 |
-
base_model:
|
| 5 |
-
- Qwen/Qwen2.5-32B-Instruct
|
| 6 |
-
pipeline_tag: text-generation
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
|
| 34 |
-
|
| 35 |
-
|
|
| 36 |
-
|
|
| 37 |
-
|
|
| 38 |
-
|
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
```
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
base_model:
|
| 5 |
+
- Qwen/Qwen2.5-32B-Instruct
|
| 6 |
+
pipeline_tag: text-generation
|
| 7 |
+
language:
|
| 8 |
+
- zho
|
| 9 |
+
- eng
|
| 10 |
+
- fra
|
| 11 |
+
- spa
|
| 12 |
+
- por
|
| 13 |
+
- deu
|
| 14 |
+
- ita
|
| 15 |
+
- rus
|
| 16 |
+
- jpn
|
| 17 |
+
- kor
|
| 18 |
+
- vie
|
| 19 |
+
- tha
|
| 20 |
+
- ara
|
| 21 |
+
---
|
| 22 |
+
|
| 23 |
+
Converted version of [Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct) to 4-bit using bitsandbytes. For more information about the model,
|
| 24 |
+
refer to the model's page.
|
| 25 |
+
|
| 26 |
+
## Impact on performance
|
| 27 |
+
Impact of quantization on a set of models.
|
| 28 |
+
|
| 29 |
+
Evaluation of the model was conducted using the PoLL (Pool of LLM) technique, assessing performance on **100 French questions** with scores aggregated from six evaluations
|
| 30 |
+
(two per evaluator). The evaluators included GPT-4o, Gemini-1.5-pro, and Claude3.5-sonnet.
|
| 31 |
+
|
| 32 |
+
Performance Scores (on a scale of 5):
|
| 33 |
+
| Model | Score | # params (Billion) | size (GB) |
|
| 34 |
+
|---------------------------------------------:|:--------:|:------------------:|:---------:|
|
| 35 |
+
| gpt-4o | 4.13 | N/A | N/A |
|
| 36 |
+
| gpt-4o-mini | 4.02 | N/A | N/A |
|
| 37 |
+
| Qwen/Qwen2.5-32B-Instruct | 3.99 | 32.8 | 65.6 |
|
| 38 |
+
| **cmarkea/Qwen2.5-32B-Instruct-4bit** | **3.98** | **32.8** | **16.4** |
|
| 39 |
+
| mistralai/Mixtral-8x7B-Instruct-v0.1 | 3.71 | 46.7 | 93.4 |
|
| 40 |
+
| cmarkea/Mixtral-8x7B-Instruct-v0.1-4bit | 3.68 | 46.7 | 23.35 |
|
| 41 |
+
| meta-llama/Meta-Llama-3.1-70B-Instruct | 3.68 | 70.06 | 140.12 |
|
| 42 |
+
| gpt-3.5-turbo | 3.66 | 175 | 350 |
|
| 43 |
+
| cmarkea/Meta-Llama-3.1-70B-Instruct-4bit | 3.64 | 70.06 | 35.3 |
|
| 44 |
+
| TheBloke/Mixtral-8x7B-Instruct-v0.1-GPTQ | 3.56 | 46.7 | 46.7 |
|
| 45 |
+
| meta-llama/Meta-Llama-3.1-8B-Instruct | 3.25 | 8.03 | 16.06 |
|
| 46 |
+
| mistralai/Mistral-7B-Instruct-v0.2 | 1.98 | 7.25 | 14.5 |
|
| 47 |
+
| cmarkea/bloomz-7b1-mt-sft-chat | 1.69 | 7.07 | 14.14 |
|
| 48 |
+
| cmarkea/bloomz-3b-dpo-chat | 1.68 | 3 | 6 |
|
| 49 |
+
| cmarkea/bloomz-3b-sft-chat | 1.51 | 3 | 6 |
|
| 50 |
+
| croissantllm/CroissantLLMChat-v0.1 | 1.19 | 1.3 | 2.7 |
|
| 51 |
+
| cmarkea/bloomz-560m-sft-chat | 1.04 | 0.56 | 1.12 |
|
| 52 |
+
| OpenLLM-France/Claire-Mistral-7B-0.1 | 0.38 | 7.25 | 14.5 |
|
| 53 |
+
|
| 54 |
+
The impact of quantization is negligible.
|
| 55 |
+
|
| 56 |
+
## Prompt Pattern
|
| 57 |
+
Here is a reminder of the command pattern to interact with the model:
|
| 58 |
+
```verbatim
|
| 59 |
+
<|im_start|>user\n{user_prompt_1}<|im_end|>\n<|im_start|>assistant\n{model_answer_1}...
|
| 60 |
```
|