Update model card
Browse files- .gitattributes +1 -0
- README.md +166 -0
- bagel.png +3 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
bagel.png filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
@@ -1,3 +1,169 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
tags:
|
3 |
+
- finetuned
|
4 |
+
- quantized
|
5 |
+
- 4-bit
|
6 |
+
- AWQ
|
7 |
+
- transformers
|
8 |
+
- pytorch
|
9 |
+
- mistral
|
10 |
+
- instruct
|
11 |
+
- text-generation
|
12 |
+
- conversational
|
13 |
+
- license:apache-2.0
|
14 |
+
- autotrain_compatible
|
15 |
+
- endpoints_compatible
|
16 |
+
- text-generation-inference
|
17 |
+
- region:us
|
18 |
+
- finetune
|
19 |
+
- chatml
|
20 |
+
- RLHF
|
21 |
+
- gpt4
|
22 |
+
- synthetic data
|
23 |
+
- distillation
|
24 |
license: apache-2.0
|
25 |
+
base_model: mistralai/mistral-7b-v0.1
|
26 |
+
datasets:
|
27 |
+
- ai2_arc
|
28 |
+
- allenai/ultrafeedback_binarized_cleaned
|
29 |
+
- argilla/distilabel-intel-orca-dpo-pairs
|
30 |
+
- jondurbin/airoboros-3.2
|
31 |
+
- codeparrot/apps
|
32 |
+
- facebook/belebele
|
33 |
+
- bluemoon-fandom-1-1-rp-cleaned
|
34 |
+
- boolq
|
35 |
+
- camel-ai/biology
|
36 |
+
- camel-ai/chemistry
|
37 |
+
- camel-ai/math
|
38 |
+
- camel-ai/physics
|
39 |
+
- jondurbin/contextual-dpo-v0.1
|
40 |
+
- jondurbin/gutenberg-dpo-v0.1
|
41 |
+
- jondurbin/py-dpo-v0.1
|
42 |
+
- jondurbin/truthy-dpo-v0.1
|
43 |
+
- LDJnr/Capybara
|
44 |
+
- jondurbin/cinematika-v0.1
|
45 |
+
- WizardLM/WizardLM_evol_instruct_70k
|
46 |
+
- glaiveai/glaive-function-calling-v2
|
47 |
+
- jondurbin/gutenberg-dpo-v0.1
|
48 |
+
- grimulkan/LimaRP-augmented
|
49 |
+
- lmsys/lmsys-chat-1m
|
50 |
+
- ParisNeo/lollms_aware_dataset
|
51 |
+
- TIGER-Lab/MathInstruct
|
52 |
+
- Muennighoff/natural-instructions
|
53 |
+
- openbookqa
|
54 |
+
- kingbri/PIPPA-shareGPT
|
55 |
+
- piqa
|
56 |
+
- Vezora/Tested-22k-Python-Alpaca
|
57 |
+
- ropes
|
58 |
+
- cakiki/rosetta-code
|
59 |
+
- Open-Orca/SlimOrca
|
60 |
+
- b-mc2/sql-create-context
|
61 |
+
- squad_v2
|
62 |
+
- mattpscott/airoboros-summarization
|
63 |
+
- migtissera/Synthia-v1.3
|
64 |
+
- unalignment/toxic-dpo-v0.2
|
65 |
+
- WhiteRabbitNeo/WRN-Chapter-1
|
66 |
+
- WhiteRabbitNeo/WRN-Chapter-2
|
67 |
+
- winogrande
|
68 |
+
language:
|
69 |
+
- en
|
70 |
+
quantized_by: Suparious
|
71 |
+
pipeline_tag: text-generation
|
72 |
+
model_creator: jondurbin
|
73 |
+
model_name: bagel 7B v0.4
|
74 |
+
inference: false
|
75 |
+
prompt_template: '{bos}<|im_start|>{role}
|
76 |
+
|
77 |
+
{text}
|
78 |
+
|
79 |
+
<|im_end|>{eos}
|
80 |
+
'
|
81 |
---
|
82 |
+
|
83 |
+
# A bagel, with everything (except DPO)
|
84 |
+
|
85 |
+
- Model creator: [jondurbin](https://huggingface.co/jondurbin)
|
86 |
+
- Original model: [bagel-7b-v0.4](https://huggingface.co/jondurbin/bagel-7b-v0.4)
|
87 |
+
|
88 |
+
![bagel](bagel.png)
|
89 |
+
|
90 |
+
## Model Description
|
91 |
+
|
92 |
+
This is the pre-DPO version of the mistral-7b model fine-tuned with https://github.com/jondurbin/bagel
|
93 |
+
|
94 |
+
The DPO counterpart will be available soon, here: https://huggingface.co/jondurbin/bagel-dpo-7b-v0.4
|
95 |
+
|
96 |
+
The non-DPO version is likely better for roleplay usage.
|
97 |
+
|
98 |
+
Compute generously provided by [MassedCompute](https://massedcompute.com/?utm_source=huggingface&utm_creative_format=model_card&utm_content=creator_jon)
|
99 |
+
|
100 |
+
## How to use
|
101 |
+
|
102 |
+
### Install the necessary packages
|
103 |
+
|
104 |
+
```bash
|
105 |
+
pip install --upgrade autoawq autoawq-kernels
|
106 |
+
```
|
107 |
+
|
108 |
+
### Example Python code
|
109 |
+
|
110 |
+
```python
|
111 |
+
from awq import AutoAWQForCausalLM
|
112 |
+
from transformers import AutoTokenizer, TextStreamer
|
113 |
+
|
114 |
+
model_path = "solidrust/bagel-7b-v0.4-AWQ"
|
115 |
+
system_message = "You are Bagel, incarnated a powerful AI with everything."
|
116 |
+
|
117 |
+
# Load model
|
118 |
+
model = AutoAWQForCausalLM.from_quantized(model_path,
|
119 |
+
fuse_layers=True)
|
120 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path,
|
121 |
+
trust_remote_code=True)
|
122 |
+
streamer = TextStreamer(tokenizer,
|
123 |
+
skip_prompt=True,
|
124 |
+
skip_special_tokens=True)
|
125 |
+
|
126 |
+
# Convert prompt to tokens
|
127 |
+
prompt_template = """\
|
128 |
+
<|im_start|>system
|
129 |
+
{system_message}<|im_end|>
|
130 |
+
<|im_start|>user
|
131 |
+
{prompt}<|im_end|>
|
132 |
+
<|im_start|>assistant"""
|
133 |
+
|
134 |
+
prompt = "You're standing on the surface of the Earth. "\
|
135 |
+
"You walk one mile south, one mile west and one mile north. "\
|
136 |
+
"You end up exactly where you started. Where are you?"
|
137 |
+
|
138 |
+
tokens = tokenizer(prompt_template.format(system_message=system_message,prompt=prompt),
|
139 |
+
return_tensors='pt').input_ids.cuda()
|
140 |
+
|
141 |
+
# Generate output
|
142 |
+
generation_output = model.generate(tokens,
|
143 |
+
streamer=streamer,
|
144 |
+
max_new_tokens=512)
|
145 |
+
```
|
146 |
+
|
147 |
+
### About AWQ
|
148 |
+
|
149 |
+
AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference with equivalent or better quality compared to the most commonly used GPTQ settings.
|
150 |
+
|
151 |
+
AWQ models are currently supported on Linux and Windows, with NVidia GPUs only. macOS users: please use GGUF models instead.
|
152 |
+
|
153 |
+
It is supported by:
|
154 |
+
|
155 |
+
- [Text Generation Webui](https://github.com/oobabooga/text-generation-webui) - using Loader: AutoAWQ
|
156 |
+
- [vLLM](https://github.com/vllm-project/vllm) - version 0.2.2 or later for support for all model types.
|
157 |
+
- [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
|
158 |
+
- [Transformers](https://huggingface.co/docs/transformers) version 4.35.0 and later, from any code or client that supports Transformers
|
159 |
+
- [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) - for use from Python code
|
160 |
+
|
161 |
+
## Prompt template: ChatML
|
162 |
+
|
163 |
+
```plaintext
|
164 |
+
<|im_start|>system
|
165 |
+
{system_message}<|im_end|>
|
166 |
+
<|im_start|>user
|
167 |
+
{prompt}<|im_end|>
|
168 |
+
<|im_start|>assistant
|
169 |
+
```
|
bagel.png
ADDED
![]() |
Git LFS Details
|