Update for Transformers GPTQ support
Browse files- README.md +25 -18
- config.json +48 -37
- gptq_model-4bit-128g.safetensors → model.safetensors +2 -2
- quantize_config.json +1 -1
README.md
CHANGED
@@ -12,17 +12,20 @@ tags:
|
|
12 |
---
|
13 |
|
14 |
<!-- header start -->
|
15 |
-
|
16 |
-
|
|
|
17 |
</div>
|
18 |
<div style="display: flex; justify-content: space-between; width: 100%;">
|
19 |
<div style="display: flex; flex-direction: column; align-items: flex-start;">
|
20 |
-
<p><a href="https://discord.gg/theblokeai">Chat & support:
|
21 |
</div>
|
22 |
<div style="display: flex; flex-direction: column; align-items: flex-end;">
|
23 |
-
<p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
|
24 |
</div>
|
25 |
</div>
|
|
|
|
|
26 |
<!-- header end -->
|
27 |
|
28 |
# NousResearch's Redmond Hermes Coder GPTQ
|
@@ -79,7 +82,7 @@ from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
|
|
79 |
import argparse
|
80 |
|
81 |
model_name_or_path = "TheBloke/Redmond-Hermes-Coder-GPTQ"
|
82 |
-
model_basename = "
|
83 |
|
84 |
use_triton = False
|
85 |
|
@@ -145,6 +148,7 @@ It was created with group_size 128 to increase inference accuracy, but without -
|
|
145 |
* Parameters: Groupsize = 128. Act Order / desc_act = False.
|
146 |
|
147 |
<!-- footer start -->
|
|
|
148 |
## Discord
|
149 |
|
150 |
For further support, and discussions on these models and AI in general, join us at:
|
@@ -164,12 +168,15 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
|
|
164 |
* Patreon: https://patreon.com/TheBlokeAI
|
165 |
* Ko-Fi: https://ko-fi.com/TheBlokeAI
|
166 |
|
167 |
-
**Special thanks to**:
|
|
|
|
|
168 |
|
169 |
-
**Patreon special mentions**: zynix , ya boyyy, Trenton Dambrowitz, Imad Khwaja, Alps Aficionado, chris gileta, John Detwiler, Willem Michiel, RoA, Mano Prime, Rainer Wilmers, Fred von Graf, Matthew Berman, Ghost , Nathan LeClaire, Iucharbius , Ai Maven, Illia Dulskyi, Joseph William Delisle, Space Cruiser, Lone Striker, Karl Bernard, Eugene Pentland, Greatston Gnanesh, Jonathan Leane, Randy H, Pierre Kircher, Willian Hasse, Stephen Murray, Alex , terasurfer , Edmond Seymore, Oscar Rangel, Luke Pendergrass, Asp the Wyvern, Junyu Yang, David Flickinger, Luke, Spiking Neurons AB, subjectnull, Pyrater, Nikolai Manek, senxiiz, Ajan Kanaga, Johann-Peter Hartmann, Artur Olbinski, Kevin Schuppel, Derek Yates, Kalila, K, Talal Aujan, Khalefa Al-Ahmad, Gabriel Puliatti, John Villwock, WelcomeToTheClub, Daniel P. Andersen, Preetika Verma, Deep Realms, Fen Risland, trip7s trip, webtim, Sean Connelly, Michael Levine, Chris McCloskey, biorpg, vamX, Viktor Bowallius, Cory Kujawski.
|
170 |
|
171 |
Thank you to all my generous patrons and donaters!
|
172 |
|
|
|
|
|
173 |
<!-- footer end -->
|
174 |
|
175 |
# Original model card: NousResearch's Redmond Hermes Coder
|
@@ -181,7 +188,7 @@ Thank you to all my generous patrons and donaters!
|
|
181 |
|
182 |
Redmond-Hermes-Coder 15B is a state-of-the-art language model fine-tuned on over 300,000 instructions. This model was fine-tuned by Nous Research, with Teknium and Karan4D leading the fine tuning process and dataset curation, Redmond AI sponsoring the compute, and several other contributors.
|
183 |
|
184 |
-
This model was trained with a WizardCoder base, which itself uses a StarCoder base model.
|
185 |
|
186 |
The model is truly great at code, but, it does come with a tradeoff though. While far better at code than the original Nous-Hermes built on Llama, it is worse than WizardCoder at pure code benchmarks, like HumanEval.
|
187 |
|
@@ -191,16 +198,16 @@ However, it does seem better at non-code than WizardCoder on a variety of things
|
|
191 |
|
192 |
## Model Training
|
193 |
|
194 |
-
The model was trained almost entirely on synthetic GPT-4 outputs. This includes data from diverse sources such as GPTeacher, the general, roleplay v1&2, code instruct datasets, Nous Instruct & PDACTL (unpublished), CodeAlpaca, Evol_Instruct Uncensored, GPT4-LLM, and Unnatural Instructions.
|
195 |
|
196 |
Additional data inputs came from Camel-AI's Biology/Physics/Chemistry and Math Datasets, Airoboros' (v1) GPT-4 Dataset, and more from CodeAlpaca. The total volume of data encompassed over 300,000 instructions.
|
197 |
|
198 |
## Collaborators
|
199 |
-
The model fine-tuning and the datasets were a collaboration of efforts and resources from members of Nous Research, includingTeknium, Karan4D, Huemin Art, and Redmond AI's generous compute grants.
|
200 |
-
|
201 |
-
Huge shoutout and acknowledgement is deserved for all the dataset creators who generously share their datasets openly.
|
202 |
|
203 |
-
Among the contributors of datasets, GPTeacher was made available by Teknium, Wizard LM by nlpxucan, and the Nous Research Instruct Dataset was provided by Karan4D and HueminArt.
|
204 |
The GPT4-LLM and Unnatural Instructions were provided by Microsoft, Airoboros dataset by jondurbin, Camel-AI datasets are from Camel-AI, and CodeAlpaca dataset by Sahil 2801.
|
205 |
If anyone was left out, please open a thread in the community tab.
|
206 |
|
@@ -213,7 +220,7 @@ The model follows the Alpaca prompt format:
|
|
213 |
### Response:
|
214 |
```
|
215 |
|
216 |
-
or
|
217 |
|
218 |
```
|
219 |
### Instruction:
|
@@ -221,11 +228,11 @@ or
|
|
221 |
### Input:
|
222 |
|
223 |
### Response:
|
224 |
-
```
|
225 |
|
226 |
## Resources for Applied Use Cases:
|
227 |
-
For an example of a back and forth chatbot using huggingface transformers and discord, check out: https://github.com/teknium1/alpaca-discord
|
228 |
-
For an example of a roleplaying discord bot, check out this: https://github.com/teknium1/alpaca-roleplay-discordbot
|
229 |
|
230 |
## Future Plans
|
231 |
The model is currently being uploaded in FP16 format, and there are plans to convert the model to GGML and GPTQ 4bit quantizations. The team is also working on a full benchmark, similar to what was done for GPT4-x-Vicuna. We will try to get in discussions to get the model included in the GPT4All.
|
@@ -270,5 +277,5 @@ HumanEval: 39%
|
|
270 |
|
271 |
## Model Usage
|
272 |
The model is available for download on Hugging Face. It is suitable for a wide range of language tasks, from generating creative text to understanding and following complex instructions.
|
273 |
-
|
274 |
Compute provided by our project sponsor Redmond AI, thank you!!
|
|
|
12 |
---
|
13 |
|
14 |
<!-- header start -->
|
15 |
+
<!-- 200823 -->
|
16 |
+
<div style="width: auto; margin-left: auto; margin-right: auto">
|
17 |
+
<img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
|
18 |
</div>
|
19 |
<div style="display: flex; justify-content: space-between; width: 100%;">
|
20 |
<div style="display: flex; flex-direction: column; align-items: flex-start;">
|
21 |
+
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
|
22 |
</div>
|
23 |
<div style="display: flex; flex-direction: column; align-items: flex-end;">
|
24 |
+
<p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
|
25 |
</div>
|
26 |
</div>
|
27 |
+
<div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
|
28 |
+
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
|
29 |
<!-- header end -->
|
30 |
|
31 |
# NousResearch's Redmond Hermes Coder GPTQ
|
|
|
82 |
import argparse
|
83 |
|
84 |
model_name_or_path = "TheBloke/Redmond-Hermes-Coder-GPTQ"
|
85 |
+
model_basename = "model"
|
86 |
|
87 |
use_triton = False
|
88 |
|
|
|
148 |
* Parameters: Groupsize = 128. Act Order / desc_act = False.
|
149 |
|
150 |
<!-- footer start -->
|
151 |
+
<!-- 200823 -->
|
152 |
## Discord
|
153 |
|
154 |
For further support, and discussions on these models and AI in general, join us at:
|
|
|
168 |
* Patreon: https://patreon.com/TheBlokeAI
|
169 |
* Ko-Fi: https://ko-fi.com/TheBlokeAI
|
170 |
|
171 |
+
**Special thanks to**: Aemon Algiz.
|
172 |
+
|
173 |
+
**Patreon special mentions**: Sam, theTransient, Jonathan Leane, Steven Wood, webtim, Johann-Peter Hartmann, Geoffrey Montalvo, Gabriel Tamborski, Willem Michiel, John Villwock, Derek Yates, Mesiah Bishop, Eugene Pentland, Pieter, Chadd, Stephen Murray, Daniel P. Andersen, terasurfer, Brandon Frisco, Thomas Belote, Sid, Nathan LeClaire, Magnesian, Alps Aficionado, Stanislav Ovsiannikov, Alex, Joseph William Delisle, Nikolai Manek, Michael Davis, Junyu Yang, K, J, Spencer Kim, Stefan Sabev, Olusegun Samson, transmissions 11, Michael Levine, Cory Kujawski, Rainer Wilmers, zynix, Kalila, Luke @flexchar, Ajan Kanaga, Mandus, vamX, Ai Maven, Mano Prime, Matthew Berman, subjectnull, Vitor Caleffi, Clay Pascal, biorpg, alfie_i, 阿明, Jeffrey Morgan, ya boyyy, Raymond Fosdick, knownsqashed, Olakabola, Leonard Tan, ReadyPlayerEmma, Enrico Ros, Dave, Talal Aujan, Illia Dulskyi, Sean Connelly, senxiiz, Artur Olbinski, Elle, Raven Klaugh, Fen Risland, Deep Realms, Imad Khwaja, Fred von Graf, Will Dee, usrbinkat, SuperWojo, Alexandros Triantafyllidis, Swaroop Kallakuri, Dan Guido, John Detwiler, Pedro Madruga, Iucharbius, Viktor Bowallius, Asp the Wyvern, Edmond Seymore, Trenton Dambrowitz, Space Cruiser, Spiking Neurons AB, Pyrater, LangChain4j, Tony Hughes, Kacper Wikieł, Rishabh Srivastava, David Ziegler, Luke Pendergrass, Andrey, Gabriel Puliatti, Lone Striker, Sebastain Graf, Pierre Kircher, Randy H, NimbleBox.ai, Vadim, danny, Deo Leter
|
174 |
|
|
|
175 |
|
176 |
Thank you to all my generous patrons and donaters!
|
177 |
|
178 |
+
And thank you again to a16z for their generous grant.
|
179 |
+
|
180 |
<!-- footer end -->
|
181 |
|
182 |
# Original model card: NousResearch's Redmond Hermes Coder
|
|
|
188 |
|
189 |
Redmond-Hermes-Coder 15B is a state-of-the-art language model fine-tuned on over 300,000 instructions. This model was fine-tuned by Nous Research, with Teknium and Karan4D leading the fine tuning process and dataset curation, Redmond AI sponsoring the compute, and several other contributors.
|
190 |
|
191 |
+
This model was trained with a WizardCoder base, which itself uses a StarCoder base model.
|
192 |
|
193 |
The model is truly great at code, but, it does come with a tradeoff though. While far better at code than the original Nous-Hermes built on Llama, it is worse than WizardCoder at pure code benchmarks, like HumanEval.
|
194 |
|
|
|
198 |
|
199 |
## Model Training
|
200 |
|
201 |
+
The model was trained almost entirely on synthetic GPT-4 outputs. This includes data from diverse sources such as GPTeacher, the general, roleplay v1&2, code instruct datasets, Nous Instruct & PDACTL (unpublished), CodeAlpaca, Evol_Instruct Uncensored, GPT4-LLM, and Unnatural Instructions.
|
202 |
|
203 |
Additional data inputs came from Camel-AI's Biology/Physics/Chemistry and Math Datasets, Airoboros' (v1) GPT-4 Dataset, and more from CodeAlpaca. The total volume of data encompassed over 300,000 instructions.
|
204 |
|
205 |
## Collaborators
|
206 |
+
The model fine-tuning and the datasets were a collaboration of efforts and resources from members of Nous Research, includingTeknium, Karan4D, Huemin Art, and Redmond AI's generous compute grants.
|
207 |
+
|
208 |
+
Huge shoutout and acknowledgement is deserved for all the dataset creators who generously share their datasets openly.
|
209 |
|
210 |
+
Among the contributors of datasets, GPTeacher was made available by Teknium, Wizard LM by nlpxucan, and the Nous Research Instruct Dataset was provided by Karan4D and HueminArt.
|
211 |
The GPT4-LLM and Unnatural Instructions were provided by Microsoft, Airoboros dataset by jondurbin, Camel-AI datasets are from Camel-AI, and CodeAlpaca dataset by Sahil 2801.
|
212 |
If anyone was left out, please open a thread in the community tab.
|
213 |
|
|
|
220 |
### Response:
|
221 |
```
|
222 |
|
223 |
+
or
|
224 |
|
225 |
```
|
226 |
### Instruction:
|
|
|
228 |
### Input:
|
229 |
|
230 |
### Response:
|
231 |
+
```
|
232 |
|
233 |
## Resources for Applied Use Cases:
|
234 |
+
For an example of a back and forth chatbot using huggingface transformers and discord, check out: https://github.com/teknium1/alpaca-discord
|
235 |
+
For an example of a roleplaying discord bot, check out this: https://github.com/teknium1/alpaca-roleplay-discordbot
|
236 |
|
237 |
## Future Plans
|
238 |
The model is currently being uploaded in FP16 format, and there are plans to convert the model to GGML and GPTQ 4bit quantizations. The team is also working on a full benchmark, similar to what was done for GPT4-x-Vicuna. We will try to get in discussions to get the model included in the GPT4All.
|
|
|
277 |
|
278 |
## Model Usage
|
279 |
The model is available for download on Hugging Face. It is suitable for a wide range of language tasks, from generating creative text to understanding and following complex instructions.
|
280 |
+
|
281 |
Compute provided by our project sponsor Redmond AI, thank you!!
|
config.json
CHANGED
@@ -1,39 +1,50 @@
|
|
1 |
{
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
}
|
|
|
1 |
{
|
2 |
+
"_name_or_path": "./hermeswizardcoder-step3800/",
|
3 |
+
"activation_function": "gelu",
|
4 |
+
"architectures": [
|
5 |
+
"GPTBigCodeForCausalLM"
|
6 |
+
],
|
7 |
+
"attention_softmax_in_fp32": true,
|
8 |
+
"attn_pdrop": 0.1,
|
9 |
+
"bos_token_id": 0,
|
10 |
+
"embd_pdrop": 0.1,
|
11 |
+
"eos_token_id": 0,
|
12 |
+
"inference_runner": 0,
|
13 |
+
"initializer_range": 0.02,
|
14 |
+
"layer_norm_epsilon": 1e-05,
|
15 |
+
"max_batch_size": null,
|
16 |
+
"max_sequence_length": null,
|
17 |
+
"model_type": "gpt_bigcode",
|
18 |
+
"multi_query": true,
|
19 |
+
"n_embd": 6144,
|
20 |
+
"n_head": 48,
|
21 |
+
"n_inner": 24576,
|
22 |
+
"n_layer": 40,
|
23 |
+
"n_positions": 8192,
|
24 |
+
"pad_key_length": true,
|
25 |
+
"pre_allocate_kv_cache": false,
|
26 |
+
"resid_pdrop": 0.1,
|
27 |
+
"scale_attention_softmax_in_fp32": true,
|
28 |
+
"scale_attn_weights": true,
|
29 |
+
"summary_activation": null,
|
30 |
+
"summary_first_dropout": 0.1,
|
31 |
+
"summary_proj_to_labels": true,
|
32 |
+
"summary_type": "cls_index",
|
33 |
+
"summary_use_proj": true,
|
34 |
+
"torch_dtype": "float16",
|
35 |
+
"transformers_version": "4.29.2",
|
36 |
+
"use_cache": true,
|
37 |
+
"validate_runner_input": true,
|
38 |
+
"vocab_size": 49153,
|
39 |
+
"quantization_config": {
|
40 |
+
"bits": 4,
|
41 |
+
"group_size": 128,
|
42 |
+
"damp_percent": 0.01,
|
43 |
+
"desc_act": false,
|
44 |
+
"sym": true,
|
45 |
+
"true_sequential": true,
|
46 |
+
"model_name_or_path": null,
|
47 |
+
"model_file_base_name": "model",
|
48 |
+
"quant_method": "gptq"
|
49 |
+
}
|
50 |
}
|
gptq_model-4bit-128g.safetensors → model.safetensors
RENAMED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ed15545d44aba2bbc238f0fae3b3f7ddf7b8b04c57a9793e61b1e02afb49bb87
|
3 |
+
size 9198428952
|
quantize_config.json
CHANGED
@@ -6,5 +6,5 @@
|
|
6 |
"sym": true,
|
7 |
"true_sequential": true,
|
8 |
"model_name_or_path": null,
|
9 |
-
"model_file_base_name":
|
10 |
}
|
|
|
6 |
"sym": true,
|
7 |
"true_sequential": true,
|
8 |
"model_name_or_path": null,
|
9 |
+
"model_file_base_name": "model"
|
10 |
}
|