Commit
·
e4a0d4d
1
Parent(s):
4d49d73
Update README.md
Browse files
README.md
CHANGED
|
@@ -90,6 +90,18 @@ Hard acc:48.03
|
|
| 90 |
|
| 91 |
**Zero-shot ACC 0.5921152388172858** (Outperforms WizardMath-7B and Qwen-7B)
|
| 92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
# 因果语言模型 7B - 与 Meta LLaMA 2 完全兼容
|
| 94 |
使用无需远程/外部代码的transformers库加载模型,AutoModelForCausalLM和AutoTokenizer(或者手动指定LlamaForCausalLM加载LM, GPT2Tokenizer加载Tokenizer),并且模型量化与GGUF(llama.cpp)、GPTQ、AWQ完全兼容。
|
| 95 |
|
|
@@ -144,4 +156,16 @@ STEM准确率:61.67
|
|
| 144 |
|
| 145 |
## GSM8K
|
| 146 |
|
| 147 |
-
**零样本准确率0.5921152388172858** (优于WizardMath-7B和Qwen-7B)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
|
| 91 |
**Zero-shot ACC 0.5921152388172858** (Outperforms WizardMath-7B and Qwen-7B)
|
| 92 |
|
| 93 |
+
## MT-Behch on DPO Version
|
| 94 |
+
| Model | MT-Bench |
|
| 95 |
+
| ------------------------- | ------------ |
|
| 96 |
+
| GPT-4 | 8.99 |
|
| 97 |
+
| GPT-3.5-Turbo | 7.94 |
|
| 98 |
+
| | |
|
| 99 |
+
| Zephyr-7b-β (Overfitting) | 7.34 |
|
| 100 |
+
| Zephyr-7b-α | 6.88 |
|
| 101 |
+
| | |
|
| 102 |
+
| **[CausalLM/14B-DPO-α](https://huggingface.co/CausalLM/14B-DPO-alpha)** | **7.618868** |
|
| 103 |
+
| **[CausalLM/7B-DPO-α](https://huggingface.co/CausalLM/7B-DPO-alpha)** | **7.038125** |
|
| 104 |
+
|
| 105 |
# 因果语言模型 7B - 与 Meta LLaMA 2 完全兼容
|
| 106 |
使用无需远程/外部代码的transformers库加载模型,AutoModelForCausalLM和AutoTokenizer(或者手动指定LlamaForCausalLM加载LM, GPT2Tokenizer加载Tokenizer),并且模型量化与GGUF(llama.cpp)、GPTQ、AWQ完全兼容。
|
| 107 |
|
|
|
|
| 156 |
|
| 157 |
## GSM8K
|
| 158 |
|
| 159 |
+
**零样本准确率0.5921152388172858** (优于WizardMath-7B和Qwen-7B)
|
| 160 |
+
|
| 161 |
+
## DPO 版本的 MT-Behch
|
| 162 |
+
| Model | MT-Bench |
|
| 163 |
+
| ------------------------- | ------------ |
|
| 164 |
+
| GPT-4 | 8.99 |
|
| 165 |
+
| GPT-3.5-Turbo | 7.94 |
|
| 166 |
+
| | |
|
| 167 |
+
| Zephyr-7b-β (Overfitting) | 7.34 |
|
| 168 |
+
| Zephyr-7b-α | 6.88 |
|
| 169 |
+
| | |
|
| 170 |
+
| **[CausalLM/14B-DPO-α](https://huggingface.co/CausalLM/14B-DPO-alpha)** | **7.618868** |
|
| 171 |
+
| **[CausalLM/7B-DPO-α](https://huggingface.co/CausalLM/7B-DPO-alpha)** | **7.038125** |
|