Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
---
|
2 |
-
base_model: HuggingFaceH4/
|
3 |
tags:
|
4 |
- alignment-handbook
|
5 |
- generated_from_trainer
|
@@ -7,16 +7,92 @@ datasets:
|
|
7 |
- HuggingFaceH4/ultrafeedback_binarized
|
8 |
- HuggingFaceH4/orca_dpo_pairs
|
9 |
model-index:
|
10 |
-
- name:
|
11 |
results: []
|
12 |
---
|
13 |
|
14 |
-
|
15 |
-
should probably proofread and complete it, then remove this comment. -->
|
16 |
|
17 |
-
#
|
18 |
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
It achieves the following results on the evaluation set:
|
21 |
- Loss: 0.4347
|
22 |
- Rewards/chosen: -0.9461
|
|
|
1 |
---
|
2 |
+
base_model: HuggingFaceH4/starchat2-15b-sft-v0.1
|
3 |
tags:
|
4 |
- alignment-handbook
|
5 |
- generated_from_trainer
|
|
|
7 |
- HuggingFaceH4/ultrafeedback_binarized
|
8 |
- HuggingFaceH4/orca_dpo_pairs
|
9 |
model-index:
|
10 |
+
- name: starchat2-15b-v0.1
|
11 |
results: []
|
12 |
---
|
13 |
|
14 |
+
<img src="https://huggingface.co/HuggingFaceH4/starchat2-15b-v0.1/resolve/main/model_logo.png" alt="StarChat2 15B Logo" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
|
|
|
15 |
|
16 |
+
# Model Card for StarChat2 15B
|
17 |
|
18 |
+
StarChat is a series of language models that are trained to act as helpful coding assistants. StarChat2 is the latest model in the series, and is a fine-tuned version of [StarCoder2](https://huggingface.co/bigcode/starcoder2-15b) that was trained with SFT and DPO on a mix of synthetic datasets.
|
19 |
+
|
20 |
+
## Model Details
|
21 |
+
|
22 |
+
### Model Description
|
23 |
+
|
24 |
+
<!-- Provide a longer summary of what this model is. -->
|
25 |
+
|
26 |
+
- **Model type:** A 16B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
|
27 |
+
- **Language(s) (NLP):** Primarily English and 80+ programming languages.
|
28 |
+
- **License:** BigCode Open RAIL-M v1
|
29 |
+
- **Finetuned from model:** [bigcode/starcoder2-15b](https://huggingface.co/bigcode/starcoder2-15b)
|
30 |
+
|
31 |
+
### Model Sources
|
32 |
+
|
33 |
+
<!-- Provide the basic links for the model. -->
|
34 |
+
|
35 |
+
- **Repository:** https://github.com/huggingface/alignment-handbook
|
36 |
+
- **Demo:** https://huggingface.co/spaces/HuggingFaceH4/starchat2-playground
|
37 |
+
|
38 |
+
|
39 |
+
## Intended uses & limitations
|
40 |
+
|
41 |
+
The model was fine-tuned on a blend of chat, code, math, and reasoning datasets. As a result, the model can be used for chat and you can check out our [demo](https://huggingface.co/spaces/HuggingFaceH4/starchat2-playground) to test its coding capabilities.
|
42 |
+
|
43 |
+
Here's how you can run the model using the `pipeline()` function from 🤗 Transformers:
|
44 |
+
|
45 |
+
```python
|
46 |
+
# pip install 'transformers @ git+https://github.com/huggingface/transformers.git@831bc25d8fdb85768402f772cf65cc3d7872b211'
|
47 |
+
# pip install accelerate
|
48 |
+
|
49 |
+
import torch
|
50 |
+
from transformers import pipeline
|
51 |
+
|
52 |
+
pipe = pipeline(
|
53 |
+
"text-generation",
|
54 |
+
model="HuggingFaceH4/starchat2-15b-v0.1",
|
55 |
+
device_map="auto",
|
56 |
+
torch_dtype=torch.bfloat16,
|
57 |
+
)
|
58 |
+
messages = [
|
59 |
+
{
|
60 |
+
"role": "system",
|
61 |
+
"content": "You are StarChat2, an expert programming assistant",
|
62 |
+
},
|
63 |
+
{"role": "user", "content": "Write a simple website in HTML. When a user clicks the button, it shows a random Chuck Norris joke."},
|
64 |
+
]
|
65 |
+
outputs = pipe(
|
66 |
+
messages,
|
67 |
+
max_new_tokens=512,
|
68 |
+
do_sample=True,
|
69 |
+
temperature=0.7,
|
70 |
+
top_k=50,
|
71 |
+
top_p=0.95,
|
72 |
+
stop_sequence="<|im_end|>",
|
73 |
+
)
|
74 |
+
print(outputs[0]["generated_text"][-1]["content"])
|
75 |
+
```
|
76 |
+
|
77 |
+
## Bias, Risks, and Limitations
|
78 |
+
|
79 |
+
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
80 |
+
|
81 |
+
StarChat2 15B has not been aligned to human preferences with techniques like RLHF or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so).
|
82 |
+
Models trained primarily on code data will also have a more skewed demographic bias commensurate with the demographics of the GitHub community, for more on this see the [StarCoder2 dataset](https://huggingface.co/datasets/bigcode/the-stack-v2)
|
83 |
+
|
84 |
+
Since the base model was pretrained on a large corpus of code, it may produce code snippets that are syntactically valid but semantically incorrect.
|
85 |
+
For example, it may produce code that does not compile or that produces incorrect results.
|
86 |
+
It may also produce code that is vulnerable to security exploits.
|
87 |
+
We have observed the model also has a tendency to produce false URLs which should be carefully inspected before clicking.
|
88 |
+
|
89 |
+
StarChat2 15B was fine-tuned from the base model [StarCoder2](https://huggingface.co/bigcode/starcoder2-15b), please refer to its model card's [Limitations Section](https://huggingface.co/bigcode/starcoder2-15b#limitations) for relevant information.
|
90 |
+
In particular, the model was evaluated on some categories of gender biases, propensity for toxicity, and risk of suggesting code completions with known security flaws; these evaluations are reported in its [technical report](https://huggingface.co/papers/2402.19173).
|
91 |
+
|
92 |
+
|
93 |
+
## Training details
|
94 |
+
|
95 |
+
This model is a fine-tuned version of [starchat2-15b-sft-v0.1](https://huggingface.co/HuggingFaceH4/starchat2-15b-sft-v0.1) on the HuggingFaceH4/ultrafeedback_binarized and the HuggingFaceH4/orca_dpo_pairs datasets.
|
96 |
It achieves the following results on the evaluation set:
|
97 |
- Loss: 0.4347
|
98 |
- Rewards/chosen: -0.9461
|