TheBloke commited on
Commit
785bd7c
1 Parent(s): 320169c

Upload new GPTQs with varied parameters

Browse files
Files changed (1) hide show
  1. README.md +68 -31
README.md CHANGED
@@ -1,10 +1,11 @@
1
  ---
2
- inference: false
3
- license: other
4
  datasets:
5
  - ehartford/WizardLM_evol_instruct_V2_196k_unfiltered_merged_split
 
6
  language:
7
  - en
 
 
8
  ---
9
 
10
  <!-- header start -->
@@ -23,52 +24,83 @@ language:
23
 
24
  # Eric Hartford's WizardLM 33B V1.0 Uncensored GPTQ
25
 
26
- These files are GPTQ 4bit model files for [Eric Hartford's WizardLM-33B-V1.0-Uncensored](https://huggingface.co/ehartford/WizardLM-33b-V1.0-Uncensored).
27
 
28
- It is the result of quantising to 4bit using [GPTQ-for-LLaMa](https://github.com/qwopqwop200/GPTQ-for-LLaMa).
 
 
29
 
30
  ## Repositories available
31
 
32
- * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ)
33
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/WizardLM-33B-V1.0-Uncensored-GGML)
34
  * [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/ehartford/WizardLM-33b-V1.0-Uncensored)
35
 
36
- ## Prompt template
37
 
38
  ```
39
- You are a helpful AI assistant.
40
 
41
- USER: <prompt>
42
  ASSISTANT:
43
  ```
44
 
45
- ## How to easily download and use this model in text-generation-webui
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
 
47
- Please make sure you're using the latest version of text-generation-webui
 
 
48
 
49
  1. Click the **Model tab**.
50
  2. Under **Download custom model or LoRA**, enter `TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ`.
 
 
51
  3. Click **Download**.
52
  4. The model will start downloading. Once it's finished it will say "Done"
53
  5. In the top left, click the refresh icon next to **Model**.
54
  6. In the **Model** dropdown, choose the model you just downloaded: `WizardLM-33B-V1.0-Uncensored-GPTQ`
55
  7. The model will automatically load, and is now ready for use!
56
  8. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.
57
- * Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
58
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
59
 
60
  ## How to use this GPTQ model from Python code
61
 
62
  First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) installed:
63
 
64
- `pip install auto-gptq`
65
 
66
  Then try the following example code:
67
 
68
  ```python
69
  from transformers import AutoTokenizer, pipeline, logging
70
  from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
71
- import argparse
72
 
73
  model_name_or_path = "TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ"
74
  model_basename = "wizardlm-33b-v1.0-uncensored-GPTQ-4bit--1g.act.order"
@@ -78,17 +110,31 @@ use_triton = False
78
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
79
 
80
  model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
81
- model_basename=model_basename,
82
  use_safetensors=True,
83
  trust_remote_code=False,
84
  device="cuda:0",
85
  use_triton=use_triton,
86
  quantize_config=None)
87
 
88
- # Note: check the prompt template is correct for this model.
 
 
 
 
 
 
 
 
 
 
 
89
  prompt = "Tell me about AI"
90
- prompt_template=f'''USER: {prompt}
91
- ASSISTANT:'''
 
 
 
92
 
93
  print("\n\n*** Generate:")
94
 
@@ -115,20 +161,11 @@ pipe = pipeline(
115
  print(pipe(prompt_template)[0]['generated_text'])
116
  ```
117
 
118
- ## Provided files
119
-
120
- **wizardlm-33b-v1.0-uncensored-GPTQ-4bit--1g.act.order.safetensors**
121
-
122
- This will work with AutoGPTQ, ExLlama, and CUDA versions of GPTQ-for-LLaMa. There are reports of issues with Triton mode of recent GPTQ-for-LLaMa. If you have issues, please use AutoGPTQ instead.
123
 
124
- It was created without group_size to lower VRAM requirements, and with --act-order (desc_act) to boost inference accuracy as much as possible.
125
 
126
- * `wizardlm-33b-v1.0-uncensored-GPTQ-4bit--1g.act.order.safetensors`
127
- * Works with AutoGPTQ in CUDA or Triton modes.
128
- * LLaMa models also work with [ExLlama](https://github.com/turboderp/exllama), which usually provides much higher performance, and uses less VRAM, than AutoGPTQ.
129
- * Works with GPTQ-for-LLaMa in CUDA mode. May have issues with GPTQ-for-LLaMa Triton mode.
130
- * Works with text-generation-webui, including one-click-installers.
131
- * Parameters: Groupsize = -1. Act Order / desc_act = True.
132
 
133
  <!-- footer start -->
134
  ## Discord
@@ -150,9 +187,9 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
150
  * Patreon: https://patreon.com/TheBlokeAI
151
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
152
 
153
- **Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
154
 
155
- **Patreon special mentions**: Pyrater, WelcomeToTheClub, Kalila, Mano Prime, Trenton Dambrowitz, Spiking Neurons AB, Pierre Kircher, Fen Risland, Kevin Schuppel, Luke, Rainer Wilmers, vamX, Gabriel Puliatti, Alex , Karl Bernard, Ajan Kanaga, Talal Aujan, Space Cruiser, ya boyyy, biorpg, Johann-Peter Hartmann, Asp the Wyvern, Ai Maven, Ghost , Preetika Verma, Nikolai Manek, trip7s trip, John Detwiler, Fred von Graf, Artur Olbinski, subjectnull, John Villwock, Junyu Yang, Rod A, Lone Striker, Chris McCloskey, Iucharbius , Matthew Berman, Illia Dulskyi, Khalefa Al-Ahmad, Imad Khwaja, chris gileta, Willem Michiel, Greatston Gnanesh, Derek Yates, K, Alps Aficionado, Oscar Rangel, David Flickinger, Luke Pendergrass, Deep Realms, Eugene Pentland, Cory Kujawski, terasurfer , Jonathan Leane, senxiiz, Joseph William Delisle, Sean Connelly, webtim, zynix , Nathan LeClaire
156
 
157
  Thank you to all my generous patrons and donaters!
158
 
 
1
  ---
 
 
2
  datasets:
3
  - ehartford/WizardLM_evol_instruct_V2_196k_unfiltered_merged_split
4
+ inference: false
5
  language:
6
  - en
7
+ license: other
8
+ model_type: llama
9
  ---
10
 
11
  <!-- header start -->
 
24
 
25
  # Eric Hartford's WizardLM 33B V1.0 Uncensored GPTQ
26
 
27
+ These files are GPTQ model files for [Eric Hartford's WizardLM 33B V1.0 Uncensored](https://huggingface.co/ehartford/WizardLM-33b-V1.0-Uncensored).
28
 
29
+ Multiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them.
30
+
31
+ These models were quantised using hardware kindly provided by [Latitude.sh](https://www.latitude.sh/accelerate).
32
 
33
  ## Repositories available
34
 
35
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ)
36
  * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/WizardLM-33B-V1.0-Uncensored-GGML)
37
  * [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/ehartford/WizardLM-33b-V1.0-Uncensored)
38
 
39
+ ## Prompt template: Vicuna
40
 
41
  ```
42
+ A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
43
 
44
+ USER: {prompt}
45
  ASSISTANT:
46
  ```
47
 
48
+ ## Provided files
49
+
50
+ Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
51
+
52
+ Each separate quant is in a different branch. See below for instructions on fetching from different branches.
53
+
54
+ | Branch | Bits | Group Size | Act Order (desc_act) | File Size | ExLlama Compatible? | Made With | Description |
55
+ | ------ | ---- | ---------- | -------------------- | --------- | ------------------- | --------- | ----------- |
56
+ | main | 4 | None | True | 16.94 GB | True | GPTQ-for-LLaMa | Most compatible option. Good inference speed in AutoGPTQ and GPTQ-for-LLaMa. Lower inference quality than other options. |
57
+ | gptq-4bit-32g-actorder_True | 4 | 32 | True | 19.44 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 32g gives highest possible inference quality, with maximum VRAM usage. Poor AutoGPTQ CUDA speed. |
58
+ | gptq-4bit-64g-actorder_True | 4 | 64 | True | 18.18 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 64g uses less VRAM than 32g, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
59
+ | gptq-4bit-128g-actorder_True | 4 | 128 | True | 17.55 GB | True | AutoGPTQ | 4-bit, with Act Order and group size. 128g uses even less VRAM, but with slightly lower accuracy. Poor AutoGPTQ CUDA speed. |
60
+ | gptq-8bit--1g-actorder_True | 8 | None | True | 32.99 GB | False | AutoGPTQ | 8-bit, with Act Order. No group size, to lower VRAM requirements and to improve AutoGPTQ speed. |
61
+ | gptq-8bit-128g-actorder_False | 8 | 128 | False | 33.73 GB | False | AutoGPTQ | 8-bit, with group size 128g for higher inference quality and without Act Order to improve AutoGPTQ speed. |
62
+ | gptq-3bit--1g-actorder_True | 3 | None | True | 12.92 GB | False | AutoGPTQ | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. |
63
+ | gptq-3bit-128g-actorder_False | 3 | 128 | False | 13.51 GB | False | AutoGPTQ | 3-bit, with group size 128g but no act-order. Slightly higher VRAM requirements than 3-bit None. |
64
+
65
+ ## How to download from branches
66
+
67
+ - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ:gptq-4bit-32g-actorder_True`
68
+ - With Git, you can clone a branch with:
69
+ ```
70
+ git clone --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ`
71
+ ```
72
+ - In Python Transformers code, the branch is the `revision` parameter; see below.
73
+
74
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
75
 
76
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
77
+
78
+ It is strongly recommended to use the text-generation-webui one-click-installers unless you know how to make a manual install.
79
 
80
  1. Click the **Model tab**.
81
  2. Under **Download custom model or LoRA**, enter `TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ`.
82
+ - To download from a specific branch, enter for example `TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ:gptq-4bit-32g-actorder_True`
83
+ - see Provided Files above for the list of branches for each option.
84
  3. Click **Download**.
85
  4. The model will start downloading. Once it's finished it will say "Done"
86
  5. In the top left, click the refresh icon next to **Model**.
87
  6. In the **Model** dropdown, choose the model you just downloaded: `WizardLM-33B-V1.0-Uncensored-GPTQ`
88
  7. The model will automatically load, and is now ready for use!
89
  8. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.
90
+ * Note that you do not need to set GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
91
  9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
92
 
93
  ## How to use this GPTQ model from Python code
94
 
95
  First make sure you have [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) installed:
96
 
97
+ `GITHUB_ACTIONS=true pip install auto-gptq`
98
 
99
  Then try the following example code:
100
 
101
  ```python
102
  from transformers import AutoTokenizer, pipeline, logging
103
  from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
 
104
 
105
  model_name_or_path = "TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ"
106
  model_basename = "wizardlm-33b-v1.0-uncensored-GPTQ-4bit--1g.act.order"
 
110
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
111
 
112
  model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
113
+ model_basename=model_basename
114
  use_safetensors=True,
115
  trust_remote_code=False,
116
  device="cuda:0",
117
  use_triton=use_triton,
118
  quantize_config=None)
119
 
120
+ """
121
+ To download from a specific branch, use the revision parameter, as in this example:
122
+
123
+ model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
124
+ revision="gptq-4bit-32g-actorder_True",
125
+ model_basename=model_basename,
126
+ use_safetensors=True,
127
+ trust_remote_code=False,
128
+ device="cuda:0",
129
+ quantize_config=None)
130
+ """
131
+
132
  prompt = "Tell me about AI"
133
+ prompt_template=f'''A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
134
+
135
+ USER: {prompt}
136
+ ASSISTANT:
137
+ '''
138
 
139
  print("\n\n*** Generate:")
140
 
 
161
  print(pipe(prompt_template)[0]['generated_text'])
162
  ```
163
 
164
+ ## Compatibility
 
 
 
 
165
 
166
+ The files provided will work with AutoGPTQ (CUDA and Triton modes), GPTQ-for-LLaMa (only CUDA has been tested), and Occ4m's GPTQ-for-LLaMa fork.
167
 
168
+ ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
 
 
 
 
 
169
 
170
  <!-- footer start -->
171
  ## Discord
 
187
  * Patreon: https://patreon.com/TheBlokeAI
188
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
189
 
190
+ **Special thanks to**: Luke from CarbonQuill, Aemon Algiz.
191
 
192
+ **Patreon special mentions**: Space Cruiser, Nikolai Manek, Sam, Chris McCloskey, Rishabh Srivastava, Kalila, Spiking Neurons AB, Khalefa Al-Ahmad, WelcomeToTheClub, Chadd, Lone Striker, Viktor Bowallius, Edmond Seymore, Ai Maven, Chris Smitley, Dave, Alexandros Triantafyllidis, Luke @flexchar, Elle, ya boyyy, Talal Aujan, Alex , Jonathan Leane, Deep Realms, Randy H, subjectnull, Preetika Verma, Joseph William Delisle, Michael Levine, chris gileta, K, Oscar Rangel, LangChain4j, Trenton Dambrowitz, Eugene Pentland, Johann-Peter Hartmann, Femi Adebogun, Illia Dulskyi, senxiiz, Daniel P. Andersen, Sean Connelly, Artur Olbinski, RoA, Mano Prime, Derek Yates, Raven Klaugh, David Flickinger, Willem Michiel, Pieter, Willian Hasse, vamX, Luke Pendergrass, webtim, Ghost , Rainer Wilmers, Nathan LeClaire, Will Dee, Cory Kujawski, John Detwiler, Fred von Graf, biorpg, Iucharbius , Imad Khwaja, Pierre Kircher, terasurfer , Asp the Wyvern, John Villwock, theTransient, zynix , Gabriel Tamborski, Fen Risland, Gabriel Puliatti, Matthew Berman, Pyrater, SuperWojo, Stephen Murray, Karl Bernard, Ajan Kanaga, Greatston Gnanesh, Junyu Yang.
193
 
194
  Thank you to all my generous patrons and donaters!
195