File tree Expand file tree Collapse file tree 1 file changed +4
-0
lines changed Expand file tree Collapse file tree 1 file changed +4
-0
lines changed Original file line number Diff line number Diff line change 39
39
40
40
41
41
## 新闻
42
+ 🔥🔥🔥 [ 2023/09/26] CodeFuse-CodeLlama-34B 4bits量化版本发布,量化后模型在HumanEval pass@1指标为73.8% (贪婪解码)。
43
+
42
44
🔥🔥🔥 [ 2023/09/07] MFTCoder微调的模型** CodeFuse-CodeLlama-34B** 在[ HumanEval Benchmarks] ( https://github.com/openai/human-eval ) 的Python ** Pass@1** 取得了** 74.4%** (greedy decoding)的开源SOTA成绩。
43
45
44
46
🔥 [ 2023/08/26] MFTCoder支持使用LoRA/QLoRA对Code Llama、Llama、Llama2、StarCoder、ChatGLM2、CodeGeeX2、Qwen和GPT-NeoX模型进行微调。
47
49
| 模型 | HumanEval(Pass@1) | 日期 |
48
50
| :----------------------------| :-----------------:| :-------:|
49
51
| ** CodeFuse-CodeLlama-34B** | ** 74.4%** | 2023/09 |
52
+ | ** CodeFuse-CodeLlama-34B-4bits** | ** 73.8%** | 2023/09 |
50
53
| WizardCoder-Python-34B-V1.0 | 73.2% | 2023/08 |
51
54
| GPT-4(zero-shot) | 67.0% | 2023/03 |
52
55
| PanGu-Coder2 15B | 61.6% | 2023/08 |
@@ -107,6 +110,7 @@ sh init_env.sh
107
110
| 模型 | 基座模型 | 训练数据 | Batch Size | Seq Length |
108
111
| ---------------------------------------------------------------| ----------------------| ------| ------------| ------------|
109
112
| [ 🔥🔥🔥 CodeFuse-CodeLlama-34B] ( https://huggingface.co/codefuse-ai/CodeFuse-CodeLlama-34B ) | CodeLlama-34b-Python | 60万 | 80 | 4096 |
113
+ | [ 🔥🔥🔥 CodeFuse-CodeLlama-34B-4bits] ( https://huggingface.co/codefuse-ai/CodeFuse-CodeLlama-34B-4bits ) | CodeLlama-34b-Python | | | 4096 |
110
114
| [ 🔥 CodeFuse-13B] ( https://huggingface.co/codefuse-ai/CodeFuse-13B ) | CodeFuse-13B-Base | 6.6万 | 64 | 4096 |
111
115
112
116
You can’t perform that action at this time.
0 commit comments