jartine commited on
Commit
9858922
1 Parent(s): 36cfc7c

Update README.md

Browse files
Files changed (1) hide show
  1. LICENSE +119 -366
LICENSE CHANGED
@@ -1,366 +1,119 @@
1
- ---
2
- base_model: microsoft/Phi-3-mini-4k-instruct
3
- license: apache-2.0
4
- language:
5
- - en
6
- pipeline_tag: text-generation
7
- tags:
8
- - llamafile
9
- prompt_template: |
10
- <|system|>
11
- You are a helpful AI assistant.<|end|>
12
- <|user|>
13
- {{prompt}}<|end|>
14
- <|assistant|>
15
- ---
16
-
17
- # Phi-3-mini-4k-instruct - llamafile
18
-
19
- This repository contains executable weights (which we call
20
- [llamafiles](https://github.com/Mozilla-Ocho/llamafile)) that run on
21
- Linux, MacOS, Windows, FreeBSD, OpenBSD, and NetBSD for AMD64 and ARM64.
22
-
23
- - Model creator: [Microsoft](https://huggingface.co/microsoft)
24
- - Original model: [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct)
25
-
26
- ## Quickstart
27
-
28
- Assuming your system has at least 32GB of RAM, you can try running the
29
- following command which download, concatenate, and execute the model.
30
-
31
- ```
32
- wget https://huggingface.co/jartine/Phi-3-mini-4k-instruct-llamafile/resolve/main/Phi-3-mini-4k-instruct.F16.llamafile
33
- chmod +x Phi-3-mini-4k-instruct.F16.llamafile
34
- ./Phi-3-mini-4k-instruct.F16.llamafile --help # view manual
35
- ./Phi-3-mini-4k-instruct.F16.llamafile # launch web gui + oai api
36
- ./Phi-3-mini-4k-instruct.F16.llamafile -p ... # cli interface (scriptable)
37
- ```
38
-
39
- Alternatively, you may download an official `llamafile` executable from
40
- Mozilla Ocho on GitHub, in which case you can use the Mixtral llamafiles
41
- as a simple weights data file.
42
-
43
- ```
44
- llamafile -m ./Phi-3-mini-4k-instruct.F16.llamafile ...
45
- ```
46
-
47
- For further information, please see the [llamafile
48
- README](https://github.com/mozilla-ocho/llamafile/).
49
-
50
- Having **trouble?** See the ["Gotchas"
51
- section](https://github.com/mozilla-ocho/llamafile/?tab=readme-ov-file#gotchas)
52
- of the README.
53
-
54
- ## Prompting
55
-
56
- Prompt template:
57
-
58
- ```
59
- <|system|>
60
- You are a helpful AI assistant.<|end|>
61
- <|user|>
62
- How to explain Internet for a medieval knight?<|end|>
63
- <|assistant|>
64
- ```
65
-
66
- Command template:
67
-
68
- ```
69
- ./Phi-3-mini-4k-instruct.F16.llamafile -e -p "<|user|>\n{{prompt}}<|end|>\n<|assistant|>"
70
- ```
71
-
72
- ## About llamafile
73
-
74
- llamafile is a new format introduced by Mozilla Ocho on Nov 20th 2023.
75
- It uses Cosmopolitan Libc to turn LLM weights into runnable llama.cpp
76
- binaries that run on the stock installs of six OSes for both ARM64 and
77
- AMD64.
78
-
79
- In addition to being executables, llamafiles are also zip archives. Each
80
- llamafile contains a GGUF file, which you can extract using the `unzip`
81
- command. If you want to change or add files to your llamafiles, then the
82
- `zipalign` command (distributed on the llamafile github) should be used
83
- instead of the traditional `zip` command.
84
-
85
- ## Licensing (Phi-3 Specific)
86
-
87
- The Phi-3 llamafiles are licensed Apache 2.0 because some of the
88
- software that went into creating these llamafiles uses that as its
89
- license. The Phi-3 weights themselves were published by Microsoft under
90
- the even more permissive MIT license. You can use the `unzip` command to
91
- extract the MIT-licensed GGUF file from each llamafile, which contains
92
- only the Microsoft Phi-3 weights.
93
-
94
- For further details on the complete picture, read our `LICENSE` file,
95
- since it documents the copyright notice of every transitive dependency.
96
-
97
- ## About Quantization Formats (General Advice)
98
-
99
- Your choice of quantization format depends on three things:
100
-
101
- 1. Will it fit in RAM or VRAM?
102
- 2. Is your use case reading (e.g. summarization) or writing (e.g. chatbot)?
103
- 3. llamafiles bigger than 4.30 GB are hard to run on Windows (see [gotchas](https://github.com/mozilla-ocho/llamafile/?tab=readme-ov-file#gotchas))
104
-
105
- Good quants for writing (prediction speed) are Q5\_K\_M, and Q4\_0. Text
106
- generation is bounded by memory speed, so smaller quants help, but they
107
- cause the LLM to hallucinate more. However that doesn't mean they can't
108
- think correctly. A highly degraded quant like `Q2_K` may not make a
109
- great encyclopedia, but it's still capable of logical reasoning and
110
- the emergent capabilities LLMs exhibit.
111
-
112
- Good quants for reading (evaluation speed) are BF16, F16, Q8\_0, and
113
- Q4\_0 (ordered from fastest to slowest). Prompt evaluation is bounded by
114
- flop count, which means perf can be improved through software
115
- engineering alone, e.g. BLAS algorithms, in which case quantization
116
- starts hurting more than it helps, since it competes for CPU resources
117
- and makes it harder for the compiler to parallelize instructions. You
118
- want to ideally use the simplest smallest floating point format that's
119
- natively implemented by your hardware. In most cases, that's BF16 or
120
- FP16. However, llamafile is able to still offer respectable tinyBLAS
121
- speedups for llama.cpp's simplest quants: Q8\_0 and Q4\_0.
122
-
123
- --
124
-
125
- ## Model Summary
126
-
127
- The Phi-3-Mini-4K-Instruct is a 3.8B parameters, lightweight, state-of-the-art open model trained with the Phi-3 datasets that includes both synthetic data and the filtered publicly available websites data with a focus on high-quality and reasoning dense properties.
128
- The model belongs to the Phi-3 family with the Mini version in two variants [4K](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) and [128K](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct) which is the context length (in tokens) that it can support.
129
-
130
- The model has underwent a post-training process that incorporates both supervised fine-tuning and direct preference optimization for the instruction following and safety measures.
131
- When assessed against benchmarks testing common sense, language understanding, math, code, long context and logical reasoning, Phi-3 Mini-4K-Instruct showcased a robust and state-of-the-art performance among models with less than 13 billion parameters.
132
-
133
- Resources and Technical Documentation:
134
-
135
- + [Phi-3 Microsoft Blog](https://aka.ms/phi3blog-april)
136
- + [Phi-3 Technical Report](https://aka.ms/phi3-tech-report)
137
- + [Phi-3 on Azure AI Studio](https://aka.ms/phi3-azure-ai)
138
- + Phi-3 GGUF: [4K](https://aka.ms/Phi3-mini-4k-instruct-gguf)
139
- + Phi-3 ONNX: [4K](https://aka.ms/Phi3-mini-4k-instruct-onnx)
140
-
141
- ## Intended Uses
142
-
143
- **Primary use cases**
144
-
145
- The model is intended for commercial and research use in English. The model provides uses for applications which require:
146
-
147
- 1) Memory/compute constrained environments
148
- 2) Latency bound scenarios
149
- 3) Strong reasoning (especially code, math and logic)
150
-
151
- Our model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
152
-
153
- **Use case considerations**
154
-
155
- Our models are not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fariness before using within a specific downstream use case, particularly for high risk scenarios. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case.
156
-
157
- Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.
158
-
159
- ## How to Use
160
-
161
- Phi-3 Mini-4K-Instruct has been integrated in the development version (4.40.0) of `transformers`. Until the official version is released through `pip`, ensure that you are doing one of the following:
162
-
163
- * When loading the model, ensure that `trust_remote_code=True` is passed as an argument of the `from_pretrained()` function.
164
-
165
- * Update your local `transformers` to the development version: `pip uninstall -y transformers && pip install git+https://github.com/huggingface/transformers`. The previous command is an alternative to cloning and installing from the source.
166
-
167
- The current `transformers` version can be verified with: `pip list | grep transformers`.
168
-
169
- Phi-3 Mini-4K-Instruct is also available in [HuggingChat](https://aka.ms/try-phi3-hf-chat).
170
-
171
- ### Tokenizer
172
-
173
- Phi-3 Mini-4K-Instruct supports a vocabulary size of up to `32064` tokens. The [tokenizer files](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/added_tokens.json) already provide placeholder tokens that can be used for downstream fine-tuning, but they can also be extended up to the model's vocabulary size.
174
-
175
- ### Chat Format
176
-
177
- Given the nature of the training data, the Phi-3 Mini-4K-Instruct model is best suited for prompts using the chat format as follows.
178
- You can provide the prompt as a question with a generic template as follow:
179
- ```markdown
180
- <|user|>\nQuestion <|end|>\n<|assistant|>
181
- ```
182
- For example:
183
- ```markdown
184
- <|system|>
185
- You are a helpful AI assistant.<|end|>
186
- <|user|>
187
- How to explain Internet for a medieval knight?<|end|>
188
- <|assistant|>
189
- ```
190
-
191
- where the model generates the text after `<|assistant|>` . In case of few-shots prompt, the prompt can be formatted as the following:
192
-
193
- ```markdown
194
- <|system|>
195
- You are a helpful AI assistant.<|end|>
196
- <|user|>
197
- I am going to Paris, what should I see?<|end|>
198
- <|assistant|>
199
- Paris, the capital of France, is known for its stunning architecture, art museums, historical landmarks, and romantic atmosphere. Here are some of the top attractions to see in Paris:\n\n1. The Eiffel Tower: The iconic Eiffel Tower is one of the most recognizable landmarks in the world and offers breathtaking views of the city.\n2. The Louvre Museum: The Louvre is one of the world's largest and most famous museums, housing an impressive collection of art and artifacts, including the Mona Lisa.\n3. Notre-Dame Cathedral: This beautiful cathedral is one of the most famous landmarks in Paris and is known for its Gothic architecture and stunning stained glass windows.\n\nThese are just a few of the many attractions that Paris has to offer. With so much to see and do, it's no wonder that Paris is one of the most popular tourist destinations in the world."<|end|>
200
- <|user|>
201
- What is so great about #1?<|end|>
202
- <|assistant|>
203
- ```
204
-
205
- ### Sample inference code
206
-
207
- This code snippets show how to get quickly started with running the model on a GPU:
208
-
209
- ```python
210
- import torch
211
- from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
212
-
213
- torch.random.manual_seed(0)
214
-
215
- model = AutoModelForCausalLM.from_pretrained(
216
- "microsoft/Phi-3-mini-4k-instruct",
217
- device_map="cuda",
218
- torch_dtype="auto",
219
- trust_remote_code=True,
220
- )
221
- tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
222
-
223
- messages = [
224
- {"role": "system", "content": "You are a helpful digital assistant. Please provide safe, ethical and accurate information to the user."},
225
- {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
226
- {"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
227
- {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
228
- ]
229
-
230
- pipe = pipeline(
231
- "text-generation",
232
- model=model,
233
- tokenizer=tokenizer,
234
- )
235
-
236
- generation_args = {
237
- "max_new_tokens": 500,
238
- "return_full_text": False,
239
- "temperature": 0.0,
240
- "do_sample": False,
241
- }
242
-
243
- output = pipe(messages, **generation_args)
244
- print(output[0]['generated_text'])
245
- ```
246
-
247
- *Some applications/frameworks might not include a BOS token (`<s>`) at the start of the conversation. Please ensure that it is included since it provides more reliable results.*
248
-
249
- ## Responsible AI Considerations
250
-
251
- Like other language models, the Phi series models can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
252
-
253
- + Quality of Service: the Phi models are trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English.
254
- + Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
255
- + Inappropriate or Offensive Content: these models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case.
256
- + Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
257
- + Limited Scope for Code: Majority of Phi-3 training data is based in Python and use common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, we strongly recommend users manually verify all API uses.
258
-
259
- Developers should apply responsible AI best practices and are responsible for ensuring that a specific use case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Important areas for consideration include:
260
-
261
- + Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
262
- + High-Risk Scenarios: Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
263
- + Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
264
- + Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
265
- + Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.
266
-
267
-
268
- ## Training
269
-
270
- ### Model
271
-
272
- * Architecture: Phi-3 Mini-4K-Instruct has 3.8B parameters and is a dense decoder-only Transformer model. The model is fine-tuned with Supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) to ensure alignment with human preferences and safety guidlines.
273
- * Inputs: Text. It is best suited for prompts using chat format.
274
- * Context length: 4K tokens
275
- * GPUs: 512 H100-80G
276
- * Training time: 7 days
277
- * Training data: 3.3T tokens
278
- * Outputs: Generated text in response to the input
279
- * Dates: Our models were trained between February and April 2024
280
- * Status: This is a static model trained on an offline dataset with cutoff date October 2023. Future versions of the tuned models may be released as we improve models.
281
-
282
- ### Datasets
283
-
284
- Our training data includes a wide variety of sources, totaling 3.3 trillion tokens, and is a combination of
285
- 1) Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code;
286
- 2) Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.);
287
- 3) High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.
288
-
289
- ### Fine-tuning
290
-
291
- A basic example of multi-GPUs supervised fine-tuning (SFT) with TRL and Accelerate modules is provided [here](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/resolve/main/sample_finetune.py).
292
-
293
- ## Benchmarks
294
-
295
- We report the results for Phi-3-Mini-4K-Instruct on standard open-source benchmarks measuring the model's reasoning ability (both common sense reasoning and logical reasoning). We compare to Phi-2, Mistral-7b-v0.1, Mixtral-8x7b, Gemma 7B, Llama-3-8B-Instruct, and GPT-3.5.
296
-
297
- All the reported numbers are produced with the exact same pipeline to ensure that the numbers are comparable. These numbers might differ from other published numbers due to slightly different choices in the evaluation.
298
-
299
- As is now standard, we use few-shot prompts to evaluate the models, at temperature 0.
300
- The prompts and number of shots are part of a Microsoft internal tool to evaluate language models, and in particular we did no optimization to the pipeline for Phi-3.
301
- More specifically, we do not change prompts, pick different few-shot examples, change prompt format, or do any other form of optimization for the model.
302
-
303
- The number of k–shot examples is listed per-benchmark.
304
-
305
- | | Phi-3-Mini-4K-In<br>3.8b | Phi-3-Small<br>7b (preview) | Phi-3-Medium<br>14b (preview) | Phi-2<br>2.7b | Mistral<br>7b | Gemma<br>7b | Llama-3-In<br>8b | Mixtral<br>8x7b | GPT-3.5<br>version 1106 |
306
- |---|---|---|---|---|---|---|---|---|---|
307
- | MMLU <br>5-Shot | 68.8 | 75.3 | 78.2 | 56.3 | 61.7 | 63.6 | 66.5 | 68.4 | 71.4 |
308
- | HellaSwag <br> 5-Shot | 76.7 | 78.7 | 83.2 | 53.6 | 58.5 | 49.8 | 71.1 | 70.4 | 78.8 |
309
- | ANLI <br> 7-Shot | 52.8 | 55.0 | 58.7 | 42.5 | 47.1 | 48.7 | 57.3 | 55.2 | 58.1 |
310
- | GSM-8K <br> 0-Shot; CoT | 82.5 | 86.4 | 90.8 | 61.1 | 46.4 | 59.8 | 77.4 | 64.7 | 78.1 |
311
- | MedQA <br> 2-Shot | 53.8 | 58.2 | 69.8 | 40.9 | 49.6 | 50.0 | 60.5 | 62.2 | 63.4 |
312
- | AGIEval <br> 0-Shot | 37.5 | 45.0 | 49.7 | 29.8 | 35.1 | 42.1 | 42.0 | 45.2 | 48.4 |
313
- | TriviaQA <br> 5-Shot | 64.0 | 59.1 | 73.3 | 45.2 | 72.3 | 75.2 | 67.7 | 82.2 | 85.8 |
314
- | Arc-C <br> 10-Shot | 84.9 | 90.7 | 91.9 | 75.9 | 78.6 | 78.3 | 82.8 | 87.3 | 87.4 |
315
- | Arc-E <br> 10-Shot | 94.6 | 97.1 | 98.0 | 88.5 | 90.6 | 91.4 | 93.4 | 95.6 | 96.3 |
316
- | PIQA <br> 5-Shot | 84.2 | 87.8 | 88.2 | 60.2 | 77.7 | 78.1 | 75.7 | 86.0 | 86.6 |
317
- | SociQA <br> 5-Shot | 76.6 | 79.0 | 79.4 | 68.3 | 74.6 | 65.5 | 73.9 | 75.9 | 68.3 |
318
- | BigBench-Hard <br> 0-Shot | 71.7 | 75.0 | 82.5 | 59.4 | 57.3 | 59.6 | 51.5 | 69.7 | 68.32 |
319
- | WinoGrande <br> 5-Shot | 70.8 | 82.5 | 81.2 | 54.7 | 54.2 | 55.6 | 65 | 62.0 | 68.8 |
320
- | OpenBookQA <br> 10-Shot | 83.2 | 88.4 | 86.6 | 73.6 | 79.8 | 78.6 | 82.6 | 85.8 | 86.0 |
321
- | BoolQ <br> 0-Shot | 77.6 | 82.9 | 86.5 | -- | 72.2 | 66.0 | 80.9 | 77.6 | 79.1 |
322
- | CommonSenseQA <br> 10-Shot | 80.2 | 80.3 | 82.6 | 69.3 | 72.6 | 76.2 | 79 | 78.1 | 79.6 |
323
- | TruthfulQA <br> 10-Shot | 65.0 | 68.1 | 74.8 | -- | 52.1 | 53.0 | 63.2 | 60.1 | 85.8 |
324
- | HumanEval <br> 0-Shot | 59.1 | 59.1 | 54.7 | 47.0 | 28.0 | 34.1 | 60.4 | 37.8 | 62.2 |
325
- | MBPP <br> 3-Shot | 53.8 | 71.4 | 73.7 | 60.6 | 50.8 | 51.5 | 67.7 | 60.2 | 77.8 |
326
-
327
- ## Software
328
-
329
- * [PyTorch](https://github.com/pytorch/pytorch)
330
- * [DeepSpeed](https://github.com/microsoft/DeepSpeed)
331
- * [Transformers](https://github.com/huggingface/transformers)
332
- * [Flash-Attention](https://github.com/HazyResearch/flash-attention)
333
-
334
- ## Hardware
335
- Note that by default, the Phi-3-mini model uses flash attention, which requires certain types of GPU hardware to run. We have tested on the following GPU types:
336
- * NVIDIA A100
337
- * NVIDIA A6000
338
- * NVIDIA H100
339
-
340
- If you want to run the model on:
341
- * NVIDIA V100 or earlier generation GPUs: call AutoModelForCausalLM.from_pretrained() with attn_implementation="eager"
342
- * CPU: use the **GGUF** quantized models [4K](https://aka.ms/Phi3-mini-4k-instruct-gguf)
343
- + Optimized inference on GPU, CPU, and Mobile: use the **ONNX** models [4K](https://aka.ms/Phi3-mini-4k-instruct-onnx)
344
-
345
-
346
- ## Cross Platform Support
347
-
348
- ONNX runtime ecosystem now supports Phi-3 Mini models across platforms and hardware. You can find the optimized Phi-3 Mini-4K-Instruct ONNX model [here](https://aka.ms/phi3-mini-4k-instruct-onnx).
349
-
350
- Optimized Phi-3 models are also published here in ONNX format, to run with ONNX Runtime on CPU and GPU across devices, including server platforms, Windows, Linux and Mac desktops, and mobile CPUs, with the precision best suited to each of these targets. DirectML support lets developers bring hardware acceleration to Windows devices at scale across AMD, Intel, and NVIDIA GPUs.
351
- Along with DirectML, ONNX Runtime provides cross platform support for Phi-3 across a range of devices CPU, GPU, and mobile.
352
-
353
- Here are some of the optimized configurations we have added:
354
-
355
- 1. ONNX models for int4 DML: Quantized to int4 via AWQ
356
- 2. ONNX model for fp16 CUDA
357
- 3. ONNX model for int4 CUDA: Quantized to int4 via RTN
358
- 4. ONNX model for int4 CPU and Mobile: Quantized to int4 via RTN
359
-
360
- ## License
361
-
362
- The model is licensed under the [MIT license](https://huggingface.co/microsoft/Phi-3-mini-4k/resolve/main/LICENSE).
363
-
364
- ## Trademarks
365
-
366
- This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.
 
1
+ Microsoft.
2
+ Copyright (c) Microsoft Corporation.
3
+
4
+ MIT License
5
+
6
+ Permission is hereby granted, free of charge, to any person obtaining a copy
7
+ of this software and associated documentation files (the "Software"), to deal
8
+ in the Software without restriction, including without limitation the rights
9
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10
+ copies of the Software, and to permit persons to whom the Software is
11
+ furnished to do so, subject to the following conditions:
12
+
13
+ The above copyright notice and this permission notice shall be included in all
14
+ copies or substantial portions of the Software.
15
+
16
+ THE SOFTWARE IS PROVIDED *AS IS*, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22
+ SOFTWARE.
23
+
24
+ llamafile (Apache 2.0)
25
+ Copyright 2023 Mozilla Foundation
26
+
27
+ Licensed under the Apache License, Version 2.0 (the "License");
28
+ you may not use this file except in compliance with the License.
29
+ You may obtain a copy of the License at
30
+
31
+ http://www.apache.org/licenses/LICENSE-2.0
32
+
33
+ Unless required by applicable law or agreed to in writing, software
34
+ distributed under the License is distributed on an "AS IS" BASIS,
35
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
36
+ See the License for the specific language governing permissions and
37
+ limitations under the License."
38
+
39
+ Cosmopolitan
40
+ Copyright 2024 Justine Alexandra Roberts Tunney
41
+
42
+ Permission to use, copy, modify, and/or distribute this software for
43
+ any purpose with or without fee is hereby granted, provided that the
44
+ above copyright notice and this permission notice appear in all copies.
45
+
46
+ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL
47
+ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED
48
+ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE
49
+ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
50
+ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
51
+ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
52
+ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
53
+ PERFORMANCE OF THIS SOFTWARE.
54
+
55
+ llama.cpp (MIT License)
56
+ Copyright (c) 2023 Georgi Gerganov
57
+ Copyright (c) 2023 Iwan Kawrakow
58
+ Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
59
+ Copyright (c) 2023 Yuji Hirose
60
+ Copyright (c) 2022 Niels Lohmann <https://nlohmann.me>
61
+ Copyright (c) 2022 Bjoern Hoehrmann <[email protected]>
62
+ Copyright (c) 2017 Sean Barrett
63
+
64
+ Permission is hereby granted, free of charge, to any person obtaining a copy
65
+ of this software and associated documentation files (the "Software"), to deal
66
+ in the Software without restriction, including without limitation the rights
67
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
68
+ copies of the Software, and to permit persons to whom the Software is
69
+ furnished to do so, subject to the following conditions:
70
+
71
+ The above copyright notice and this permission notice shall be included in all
72
+ copies or substantial portions of the Software.
73
+
74
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
75
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
76
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
77
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
78
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
79
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
80
+ SOFTWARE.
81
+
82
+ strftime (BSD-3)
83
+ Copyright 1989 The Regents of the University of California
84
+
85
+ *NSYNC (Apache 2.0)
86
+ Copyright 2016 Google, Inc.
87
+ https://github.com/google/nsync
88
+
89
+ puff (zlib License)
90
+ Copyright 2002-2013 Mark Adler
91
+
92
+ Smoothsort (MIT License)
93
+ Copyright 2011 Valentin Ochs
94
+ Discovered by Edsger Dijkstra
95
+
96
+ OpenBSD Strings (ISC)
97
+ Copyright (c) 1998, 2015 Todd C. Miller <[email protected]>
98
+
99
+ Optimized Routines (MIT License)
100
+ Copyright 2022 ARM Limited
101
+
102
+ FreeBSD libm (BSD-2 License)
103
+ Copyright (c) 1992-2024 The FreeBSD Project.
104
+ Copyright (c) 2005-2011 Bruce D. Evans, Steven G. Kargl, David Schultz
105
+ Copyright (c) 2009-2013 Steven G. Kargl
106
+
107
+ fdlibm (fdlibm license)
108
+ Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
109
+
110
+ Musl libc (MIT License)
111
+ Copyright 2005-2014 Rich Felker, et. al.
112
+
113
+ gdtoa (MIT License)
114
+ The author of this software is David M. Gay
115
+ Kudos go to Guy L. Steele, Jr. and Jon L. White
116
+ Copyright (C) 1997, 1998, 2000 by Lucent Technologies
117
+
118
+ OpenBSD Sorting (BSD-3)
119
+ Copyright 1993 The Regents of the University of California