TheBloke commited on
Commit
068e0d5
1 Parent(s): 38ea47a

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +382 -0
README.md ADDED
@@ -0,0 +1,382 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Xwin-LM/XwinCoder-13B
3
+ inference: false
4
+ license: llama2
5
+ model_creator: Xwin-LM
6
+ model_name: XwinCoder 13B
7
+ model_type: llama
8
+ prompt_template: "<system>: You are an AI coding assistant that helps people with\
9
+ \ programming. Write a response that appropriately completes the user's request.\n\
10
+ <user>: {prompt}\n<AI>: \n"
11
+ quantized_by: TheBloke
12
+ ---
13
+ <!-- markdownlint-disable MD041 -->
14
+
15
+ <!-- header start -->
16
+ <!-- 200823 -->
17
+ <div style="width: auto; margin-left: auto; margin-right: auto">
18
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
19
+ </div>
20
+ <div style="display: flex; justify-content: space-between; width: 100%;">
21
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
22
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
23
+ </div>
24
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
25
+ <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>
26
+ </div>
27
+ </div>
28
+ <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>
29
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
30
+ <!-- header end -->
31
+
32
+ # XwinCoder 13B - AWQ
33
+ - Model creator: [Xwin-LM](https://huggingface.co/Xwin-LM)
34
+ - Original model: [XwinCoder 13B](https://huggingface.co/Xwin-LM/XwinCoder-13B)
35
+
36
+ <!-- description start -->
37
+ ## Description
38
+
39
+ This repo contains AWQ model files for [Xwin-LM's XwinCoder 13B](https://huggingface.co/Xwin-LM/XwinCoder-13B).
40
+
41
+ These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
42
+
43
+
44
+ ### About AWQ
45
+
46
+ 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.
47
+
48
+ It is supported by:
49
+
50
+ - [Text Generation Webui](https://github.com/oobabooga/text-generation-webui) - using Loader: AutoAWQ
51
+ - [vLLM](https://github.com/vllm-project/vllm) - Llama and Mistral models only
52
+ - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
53
+ - [Transformers](https://huggingface.co/docs/transformers) version 4.35.0 and later, from any code or client that supports Transformers
54
+ - [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) - for use from Python code
55
+
56
+ <!-- description end -->
57
+ <!-- repositories-available start -->
58
+ ## Repositories available
59
+
60
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/XwinCoder-13B-AWQ)
61
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/XwinCoder-13B-GPTQ)
62
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/XwinCoder-13B-GGUF)
63
+ * [Xwin-LM's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/Xwin-LM/XwinCoder-13B)
64
+ <!-- repositories-available end -->
65
+
66
+ <!-- prompt-template start -->
67
+ ## Prompt template: XWin-Coder
68
+
69
+ ```
70
+ <system>: You are an AI coding assistant that helps people with programming. Write a response that appropriately completes the user's request.
71
+ <user>: {prompt}
72
+ <AI>:
73
+
74
+ ```
75
+
76
+ <!-- prompt-template end -->
77
+
78
+
79
+ <!-- README_AWQ.md-provided-files start -->
80
+ ## Provided files, and AWQ parameters
81
+
82
+ I currently release 128g GEMM models only. The addition of group_size 32 models, and GEMV kernel models, is being actively considered.
83
+
84
+ Models are released as sharded safetensors files.
85
+
86
+ | Branch | Bits | GS | AWQ Dataset | Seq Len | Size |
87
+ | ------ | ---- | -- | ----------- | ------- | ---- |
88
+ | [main](https://huggingface.co/TheBloke/XwinCoder-13B-AWQ/tree/main) | 4 | 128 | [code](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1/viewer/) | 4096 | 7.25 GB
89
+
90
+ <!-- README_AWQ.md-provided-files end -->
91
+
92
+ <!-- README_AWQ.md-text-generation-webui start -->
93
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
94
+
95
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
96
+
97
+ It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.
98
+
99
+ 1. Click the **Model tab**.
100
+ 2. Under **Download custom model or LoRA**, enter `TheBloke/XwinCoder-13B-AWQ`.
101
+ 3. Click **Download**.
102
+ 4. The model will start downloading. Once it's finished it will say "Done".
103
+ 5. In the top left, click the refresh icon next to **Model**.
104
+ 6. In the **Model** dropdown, choose the model you just downloaded: `XwinCoder-13B-AWQ`
105
+ 7. Select **Loader: AutoAWQ**.
106
+ 8. Click Load, and the model will load and is now ready for use.
107
+ 9. 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.
108
+ 10. Once you're ready, click the **Text Generation** tab and enter a prompt to get started!
109
+ <!-- README_AWQ.md-text-generation-webui end -->
110
+
111
+ <!-- README_AWQ.md-use-from-vllm start -->
112
+ ## Multi-user inference server: vLLM
113
+
114
+ Documentation on installing and using vLLM [can be found here](https://vllm.readthedocs.io/en/latest/).
115
+
116
+ - Please ensure you are using vLLM version 0.2 or later.
117
+ - When using vLLM as a server, pass the `--quantization awq` parameter.
118
+
119
+ For example:
120
+
121
+ ```shell
122
+ python3 -m vllm.entrypoints.api_server --model TheBloke/XwinCoder-13B-AWQ --quantization awq --dtype auto
123
+ ```
124
+
125
+ - When using vLLM from Python code, again set `quantization=awq`.
126
+
127
+ For example:
128
+
129
+ ```python
130
+ from vllm import LLM, SamplingParams
131
+
132
+ prompts = [
133
+ "Tell me about AI",
134
+ "Write a story about llamas",
135
+ "What is 291 - 150?",
136
+ "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
137
+ ]
138
+ prompt_template=f'''<system>: You are an AI coding assistant that helps people with programming. Write a response that appropriately completes the user's request.
139
+ <user>: {prompt}
140
+ <AI>:
141
+ '''
142
+
143
+ prompts = [prompt_template.format(prompt=prompt) for prompt in prompts]
144
+
145
+ sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
146
+
147
+ llm = LLM(model="TheBloke/XwinCoder-13B-AWQ", quantization="awq", dtype="auto")
148
+
149
+ outputs = llm.generate(prompts, sampling_params)
150
+
151
+ # Print the outputs.
152
+ for output in outputs:
153
+ prompt = output.prompt
154
+ generated_text = output.outputs[0].text
155
+ print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
156
+ ```
157
+ <!-- README_AWQ.md-use-from-vllm start -->
158
+
159
+ <!-- README_AWQ.md-use-from-tgi start -->
160
+ ## Multi-user inference server: Hugging Face Text Generation Inference (TGI)
161
+
162
+ Use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`
163
+
164
+ Example Docker parameters:
165
+
166
+ ```shell
167
+ --model-id TheBloke/XwinCoder-13B-AWQ --port 3000 --quantize awq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
168
+ ```
169
+
170
+ Example Python code for interfacing with TGI (requires [huggingface-hub](https://github.com/huggingface/huggingface_hub) 0.17.0 or later):
171
+
172
+ ```shell
173
+ pip3 install huggingface-hub
174
+ ```
175
+
176
+ ```python
177
+ from huggingface_hub import InferenceClient
178
+
179
+ endpoint_url = "https://your-endpoint-url-here"
180
+
181
+ prompt = "Tell me about AI"
182
+ prompt_template=f'''<system>: You are an AI coding assistant that helps people with programming. Write a response that appropriately completes the user's request.
183
+ <user>: {prompt}
184
+ <AI>:
185
+ '''
186
+
187
+ client = InferenceClient(endpoint_url)
188
+ response = client.text_generation(prompt,
189
+ max_new_tokens=128,
190
+ do_sample=True,
191
+ temperature=0.7,
192
+ top_p=0.95,
193
+ top_k=40,
194
+ repetition_penalty=1.1)
195
+
196
+ print(f"Model output: ", response)
197
+ ```
198
+ <!-- README_AWQ.md-use-from-tgi end -->
199
+
200
+ <!-- README_AWQ.md-use-from-python start -->
201
+ ## Inference from Python code using Transformers
202
+
203
+ ### Install the necessary packages
204
+
205
+ - Requires: [Transformers](https://huggingface.co/docs/transformers) 4.35.0 or later.
206
+ - Requires: [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) 0.1.6 or later.
207
+
208
+ ```shell
209
+ pip3 install --upgrade "autoawq>=0.1.6" "transformers>=4.35.0"
210
+ ```
211
+
212
+ Note that if you are using PyTorch 2.0.1, the above AutoAWQ command will automatically upgrade you to PyTorch 2.1.0.
213
+
214
+ If you are using CUDA 11.8 and wish to continue using PyTorch 2.0.1, instead run this command:
215
+
216
+ ```shell
217
+ pip3 install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.6/autoawq-0.1.6+cu118-cp310-cp310-linux_x86_64.whl
218
+ ```
219
+
220
+ If you have problems installing [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) using the pre-built wheels, install it from source instead:
221
+
222
+ ```shell
223
+ pip3 uninstall -y autoawq
224
+ git clone https://github.com/casper-hansen/AutoAWQ
225
+ cd AutoAWQ
226
+ pip3 install .
227
+ ```
228
+
229
+ ### Transformers example code (requires Transformers 4.35.0 and later)
230
+
231
+ ```python
232
+ from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
233
+
234
+ model_name_or_path = "TheBloke/XwinCoder-13B-AWQ"
235
+
236
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
237
+ model = AutoModelForCausalLM.from_pretrained(
238
+ model_name_or_path,
239
+ low_cpu_mem_usage=True,
240
+ device_map="cuda:0"
241
+ )
242
+
243
+ # Using the text streamer to stream output one token at a time
244
+ streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
245
+
246
+ prompt = "Tell me about AI"
247
+ prompt_template=f'''<system>: You are an AI coding assistant that helps people with programming. Write a response that appropriately completes the user's request.
248
+ <user>: {prompt}
249
+ <AI>:
250
+ '''
251
+
252
+ # Convert prompt to tokens
253
+ tokens = tokenizer(
254
+ prompt_template,
255
+ return_tensors='pt'
256
+ ).input_ids.cuda()
257
+
258
+ generation_params = {
259
+ "do_sample": True,
260
+ "temperature": 0.7,
261
+ "top_p": 0.95,
262
+ "top_k": 40,
263
+ "max_new_tokens": 512,
264
+ "repetition_penalty": 1.1
265
+ }
266
+
267
+ # Generate streamed output, visible one token at a time
268
+ generation_output = model.generate(
269
+ tokens,
270
+ streamer=streamer,
271
+ **generation_params
272
+ )
273
+
274
+ # Generation without a streamer, which will include the prompt in the output
275
+ generation_output = model.generate(
276
+ tokens,
277
+ **generation_params
278
+ )
279
+
280
+ # Get the tokens from the output, decode them, print them
281
+ token_output = generation_output[0]
282
+ text_output = tokenizer.decode(token_output)
283
+ print("model.generate output: ", text_output)
284
+
285
+ # Inference is also possible via Transformers' pipeline
286
+ from transformers import pipeline
287
+
288
+ pipe = pipeline(
289
+ "text-generation",
290
+ model=model,
291
+ tokenizer=tokenizer,
292
+ **generation_params
293
+ )
294
+
295
+ pipe_output = pipe(prompt_template)[0]['generated_text']
296
+ print("pipeline output: ", pipe_output)
297
+
298
+ ```
299
+ <!-- README_AWQ.md-use-from-python end -->
300
+
301
+ <!-- README_AWQ.md-compatibility start -->
302
+ ## Compatibility
303
+
304
+ The files provided are tested to work with:
305
+
306
+ - [text-generation-webui](https://github.com/oobabooga/text-generation-webui) using `Loader: AutoAWQ`.
307
+ - [vLLM](https://github.com/vllm-project/vllm) version 0.2.0 and later.
308
+ - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) version 1.1.0 and later.
309
+ - [Transformers](https://huggingface.co/docs/transformers) version 4.35.0 and later.
310
+ - [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) version 0.1.1 and later.
311
+
312
+ <!-- README_AWQ.md-compatibility end -->
313
+
314
+ <!-- footer start -->
315
+ <!-- 200823 -->
316
+ ## Discord
317
+
318
+ For further support, and discussions on these models and AI in general, join us at:
319
+
320
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
321
+
322
+ ## Thanks, and how to contribute
323
+
324
+ Thanks to the [chirper.ai](https://chirper.ai) team!
325
+
326
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
327
+
328
+ I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
329
+
330
+ If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
331
+
332
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
333
+
334
+ * Patreon: https://patreon.com/TheBlokeAI
335
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
336
+
337
+ **Special thanks to**: Aemon Algiz.
338
+
339
+ **Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
340
+
341
+
342
+ Thank you to all my generous patrons and donaters!
343
+
344
+ And thank you again to a16z for their generous grant.
345
+
346
+ <!-- footer end -->
347
+
348
+ # Original model card: Xwin-LM's XwinCoder 13B
349
+
350
+ # XwinCoder
351
+
352
+ We are glad to introduce our instruction finetuned code generation models based on CodeLLaMA: XwinCoder. We release model weights and evaluation code.
353
+
354
+ **Repository:** [https://github.com/Xwin-LM/Xwin-LM/tree/main/Xwin-Coder](https://github.com/Xwin-LM/Xwin-LM/tree/main/Xwin-Coder)
355
+
356
+ **Models:**
357
+ | Model | 🤗hf link | HumanEval pass@1 | MBPP pass@1 | APPS-intro pass@5 |
358
+ |-------|------------|----------|------|-------------|
359
+ | XwinCoder-7B | [link](https://huggingface.co/Xwin-LM/XwinCoder-7B) | 63.8 | 57.4 | 31.5 |
360
+ | XwinCoder-13B | [link](https://huggingface.co/Xwin-LM/XwinCoder-13B) | 68.8 | 60.1 | 35.4 |
361
+ | XwinCoder-34B | [link](https://huggingface.co/Xwin-LM/XwinCoder-34B) | 74.2 | 64.8 | 43.0 |
362
+
363
+ ## Updates
364
+ - 💥 We released [**XwinCoder-7B**](https://huggingface.co/Xwin-LM/XwinCoder-7B), [**XwinCoder-13B**](https://huggingface.co/Xwin-LM/XwinCoder-13B), [**XwinCoder-34B**](https://huggingface.co/Xwin-LM/XwinCoder-34B). Our XwinCoder-34B reached 74.2 on HumanEval and it **achieves comparable performance as GPT-3.5-turbo on 6 benchmarks**.
365
+
366
+ - ❗We support evaluating instruction finetuned models on HumanEval, MBPP, APPS, DS1000 and MT-Bench. See our github repository.
367
+ -
368
+ ## Overview
369
+
370
+ ![Chat demo](rader.png)
371
+
372
+ * To fully demonstrate our model's coding capabilities in real-world usage scenarios, we have conducted thorough evaluations on several existing mainstream coding capability leaderboards (rather than only on the currently most popular HumanEval).
373
+ * As shown in the radar chart results, our 34B model **achieves comparable performance as GPT-3.5-turbo on coding abilities**.
374
+ * It is worth mentioning that, to ensure accurate visualization, our radar chart has not been scaled (only translated; MT-Bench score is scaled by 10x to be more comparable with other benchmarks).
375
+ * Multiple-E-avg6 refer to the 6 languages used in CodeLLaMA paper. Results of GPT-4 and GPT-3.5-turbo are conducted by us, more details will be released later.
376
+
377
+ ## Demo
378
+ We provide a chat demo in our github repository, here are some examples:
379
+ ![Chat demo](exm.gif)
380
+
381
+
382
+