munish0838
commited on
Commit
•
13ac8d9
1
Parent(s):
f08d23c
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- openbmb/UltraInteract_sft
|
5 |
+
- stingning/ultrachat
|
6 |
+
- openchat/openchat_sharegpt4_dataset
|
7 |
+
- Open-Orca/OpenOrca
|
8 |
+
tags:
|
9 |
+
- reasoning
|
10 |
+
pipeline_tag: text-generation
|
11 |
+
---
|
12 |
+
|
13 |
+
# Eurus-7b-sft-GGUF
|
14 |
+
|
15 |
+
- This is quantized version of [openbmb/Eurus-7b-sft](https://huggingface.co/openbmb/Eurus-7b-sft) created using llama.cpp
|
16 |
+
|
17 |
+
# Model Description
|
18 |
+
|
19 |
+
Eurus-7B-SFT is fine-tuned from Mistral-7B on all correct actions in UltraInteract, mixing a small proportion of UltraChat, ShareGPT, and OpenOrca examples.
|
20 |
+
|
21 |
+
It achieves better performance than other open-source models of similar sizes and even outperforms specialized models in corresponding domains in many cases.
|
22 |
+
|
23 |
+
## Usage
|
24 |
+
|
25 |
+
We apply tailored prompts for coding and math, consistent with UltraInteract data formats:
|
26 |
+
|
27 |
+
**Coding**
|
28 |
+
|
29 |
+
```
|
30 |
+
[INST] Write Python code to solve the task:
|
31 |
+
{Instruction} [/INST]
|
32 |
+
```
|
33 |
+
**Math-CoT**
|
34 |
+
|
35 |
+
```
|
36 |
+
[INST] Solve the following math problem step-by-step.
|
37 |
+
Simplify your answer as much as possible. Present your final answer as \\boxed{Your Answer}.
|
38 |
+
{Instruction} [/INST]
|
39 |
+
```
|
40 |
+
|
41 |
+
**Math-PoT**
|
42 |
+
|
43 |
+
```
|
44 |
+
[INST] Tool available:
|
45 |
+
[1] Python interpreter
|
46 |
+
When you send a message containing Python code to python, it will be executed in a stateful Jupyter notebook environment.
|
47 |
+
Solve the following math problem step-by-step.
|
48 |
+
Simplify your answer as much as possible.
|
49 |
+
{Instruction} [/INST]
|
50 |
+
```
|
51 |
+
|
52 |
+
## Evaluation
|
53 |
+
- Eurus, both the 7B and 70B variants, achieve the best overall performance among open-source models of similar sizes. Eurus even outperforms specialized models in corresponding domains in many cases. Notably, Eurus-7B outperforms baselines that are 5× larger, and Eurus-70B achieves better performance than GPT-3.5 Turbo.
|
54 |
+
- Preference learning with UltraInteract can further improve performance, especially in math and the multi-turn ability.
|
55 |
+
<img src="main_exp.png" alt="stats" style="zoom: 40%;" />
|
56 |
+
|