Update README.md
Browse files
README.md
CHANGED
|
@@ -1,24 +1,26 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
|
| 12 |
-
|
|
| 13 |
-
|
|
| 14 |
-
|
|
| 15 |
-
|
|
| 16 |
-
|
|
| 17 |
-
|
|
| 18 |
-
|
|
| 19 |
-
| **
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: mistralai/Mistral-7B-Instruct-v0.3
|
| 4 |
+
---
|
| 5 |
+
# [Mistral-7B-Instruct-v0.3](mistralai/Mistral-7B-Instruct-v0.3) quantized to 4bits
|
| 6 |
+
|
| 7 |
+
- weight-only quantization via GPTQ to 4bits with group_size=128
|
| 8 |
+
- GPTQ optimized for 99.75% accuracy recovery relative to the unquantized model
|
| 9 |
+
|
| 10 |
+
# Open LLM Leaderboard evaluation scores
|
| 11 |
+
| | Mistral-7B-Instruct-v0.3 | Mistral-7B-Instruct-v0.3-GPTQ-4bit<br>(this model) |
|
| 12 |
+
| :------------------: | :----------------------: | :------------------------------------------------: |
|
| 13 |
+
| arc-c<br>25-shot | 63.48 | 63.40 |
|
| 14 |
+
| mmlu<br>5-shot | 61.13 | 60.89 |
|
| 15 |
+
| hellaswag<br>10-shot | 84.49 | 84.04 |
|
| 16 |
+
| winogrande<br>5-shot | 79.16 | 79.08 |
|
| 17 |
+
| gsm8k<br>5-shot | 43.37 | 45.41 |
|
| 18 |
+
| truthfulqa<br>0-shot | 59.65 | 57.48 |
|
| 19 |
+
| **Average<br>Accuracy** | **65.21** | **65.05** |
|
| 20 |
+
| **Recovery** | **100%** | **99.75%** |
|
| 21 |
+
|
| 22 |
+
# vLLM Inference Performance
|
| 23 |
+
|
| 24 |
+
This model is ready for optimized inference using the Marlin mixed-precision kernels in vLLM: https://github.com/vllm-project/vllm
|
| 25 |
+
|
| 26 |
+
|