mlx-lm-lora 0.1.1__tar.gz

mlx_lm_lora-0.1.1/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

mlx_lm_lora-0.1.1/MANIFEST.in

@@ -0,0 +1,2 @@
+include requirements.txt
+recursive-include mlx_lm_lora/ *.py

mlx_lm_lora-0.1.1/PKG-INFO

@@ -0,0 +1,459 @@
+Metadata-Version: 2.4
+Name: mlx-lm-lora
+Version: 0.1.1
+Summary: Train LLMs on Apple silicon with MLX and the Hugging Face Hub
+Home-page: https://github.com/Goekdeniz-Guelmez/mlx-lm-lora.git
+Author: Gökdeniz Gülmez
+Author-email: goekdenizguelmez@gmail.com
+License: MIT
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: mlx>=0.25.0
+Requires-Dist: mlx_lm
+Requires-Dist: numpy
+Requires-Dist: transformers[sentencepiece]>=4.39.3
+Requires-Dist: protobuf
+Requires-Dist: pyyaml
+Requires-Dist: jinja2
+Requires-Dist: tqdm
+Requires-Dist: wandb
+Requires-Dist: datasets
+Provides-Extra: evaluate
+Requires-Dist: lm-eval; extra == "evaluate"
+Requires-Dist: tqdm; extra == "evaluate"
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: license
+Dynamic: license-file
+Dynamic: provides-extra
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+# mlx-lm-lora
+Train Large Language Models localy on Apple Silicon using MLX. Fine-tuning works with all the model that are supported with MLX-LM, for example:
+
+- Llama, 3, 4
+- Phi2, 3
+- Mistral
+- Mixtral
+- Qwen2, 2.5, 3
+- Qwen3 MoE
+- Gemma1, 2, 3
+- OLMo, OLMoE
+- MiniCPM, MiniCPM3
+- and more ...
+
+## Contents
+
+- [Run](#Run)
+  - [LoRA or Full-Precision](#Lora-or-Full-Precision)
+  - [SFT](#SFT-Training)
+  - [ORPO-Training](#ORPO-Training)
+  - [DPO-Training](#DPO-Training)
+  - [GRPO-Training](#GRPO-Training)
+  - [Evaluate](#Evaluate)
+  - [Generate](#Generate)
+- [Fuse](#Fuse)
+- [Memory Issues](#Memory-Issues)
+
+--
+
+## Run
+
+The main command is `mlx_lm_lora.train`. To see a full list of command-line options run:
+
+```shell
+mlx_lm_lora.train --help
+```
+
+Note, in the following the `--model` argument can be any compatible Hugging
+Face repo or a local path to a converted model.
+
+You can also specify a YAML config with `-c`/`--config`. For more on the format see the
+[example YAML](examples/lora_config.yaml). For example:
+
+```shell
+mlx_lm_lora.train --config /path/to/config.yaml
+```
+
+If command-line flags are also used, they will override the corresponding
+values in the config.
+
+---
+
+### LoRA or Full-Precision
+
+To fine-tune a model use:
+
+```shell
+mlx_lm_lora.train \
+    --model <path_to_model> \
+    --train \
+    --data <path_to_data> \
+    --iters 600
+```
+
+To fine-tune the full model weights, add the `--train-type full` flag.
+Currently supported training types are `lora` (default), `dora`, and `full`.
+
+The `--data` argument must specify a path to a `train.jsonl`, `valid.jsonl`
+when using `--train` and a path to a `test.jsonl` when using `--test`. For more
+
+If `--model` points to a quantized model, then the training will use QLoRA,
+otherwise it will use regular LoRA.
+
+By default, the adapter config and learned weights are saved in `adapters/`.
+You can specify the output location with `--adapter-path`.
+
+You can resume fine-tuning with an existing adapter with
+`--resume-adapter-file <path_to_adapters.safetensors>`.
+
+---
+
+### SFT-Training
+
+Supervised Fine-Tuning (SFT) trains a model using pairs of prompts and expected completions. This is the most common form of instruction tuning.
+
+To run SFT:
+
+mlx_lm_lora.train \
+    --model <path_to_model> \
+    --train \
+    --data <path_to_data>
+
+You can set the training type explicitly using --train-type lora, dora, or full. By default, LoRA is used.
+
+Data Format
+
+The data should be in JSONL format with one of the following structures:
+
+Chat-style (preferred for chat models):
+
+{
+  "messages": [
+    {"role": "system", "content": "You are a helpful assistant."},
+    {"role": "user", "content": "What is the capital of France?"},
+    {"role": "assistant", "content": "Paris."}
+  ]
+}
+
+Prompt-completion style:
+
+{"prompt": "What is the capital of France?", "completion": "Paris."}
+
+You can mix system messages or use multi-turn chat formatting depending on the target model’s capabilities.
+
+Additional Options
+	•	--use-chat-template: Format messages using the model’s chat template (default: False)
+	•	--mask-prompt: Apply loss only on the assistant’s output instead of the full sequence
+	•	--train-type: Choose lora, dora, or full for LoRA, DoRA, or full-weight fine-tuning
+	•	--resume-adapter-file: Resume training from a previously saved adapter
+
+Example with chat template and prompt masking:
+
+mlx_lm_lora.train \
+    --model <path_to_model> \
+    --train \
+    --data <path_to_data> \
+    --use-chat-template True \
+    --mask-prompt
+
+---
+
+### ORPO-Training
+
+Odds Ratio Preference Optimization (ORPO) training fine-tunes models using human preference data. Usage:
+
+```shell
+mlx_lm_lora.train \
+ --model <path_to_model> \
+ --train \
+ --train-mode orpo \
+ --data <path_to_data> \
+ --beta 0.1
+```
+
+Parameters:
+
+- `--beta`: Temperature for logistic function (default: 0.1)
+
+Data format (JSONL):
+
+```jsonl
+# Basic format with string responses
+{"prompt": "User prompt", "chosen": "Preferred response", "rejected": "Less preferred response"}
+
+# With custom preference score
+{"prompt": "User prompt", "chosen": "Preferred response", "rejected": "Less preferred response", "preference_score": 8.0}
+
+# With system message
+{"prompt": "User prompt", "chosen": "Preferred response", "rejected": "Less preferred response", "system": "System instruction"}
+
+# With full conversation objects
+{
+  "prompt": "User prompt",
+  "chosen": {
+    "messages": [
+      {"role": "system", "content": "System instruction"},
+      {"role": "user", "content": "User message"},
+      {"role": "assistant", "content": "Assistant response"}
+    ]
+  },
+  "rejected": {
+    "messages": [
+      {"role": "system", "content": "System instruction"},
+      {"role": "user", "content": "User message"},
+      {"role": "assistant", "content": "Assistant response"}
+    ]
+  }
+}
+```
+
+The trainer assigns binary rewards (1.0 chosen, 0.0 rejected) if no explicit rewards provided via `preference_score`.
+
+---
+
+### DPO-Training
+
+Direct Preference Optimization (DPO) training allows you to fine-tune models using human preference data. To use DPO training, set the training mode to 'dpo':
+
+```shell
+mlx_lm.lora \
+    --model <path_to_model> \
+    --train \
+    --train-mode dpo \
+    --data <path_to_data> \
+    --beta 0.1
+```
+
+The DPO training accepts the following additional parameters:
+
+- `--beta`: Controls the strength of the DPO loss (default: 0.1)
+- `--dpo-loss-type`: Choose between "sigmoid" (default), "hinge", "ipo", or "dpop" loss functions
+- `--delta`: Margin parameter for hinge loss (default: 50.0)
+- `--reference-model-path`: Path to a reference model for DPO training
+
+For DPO training, the data should be in JSONL format with the following structure:
+
+```jsonl
+{"prompt": "User prompt", "chosen": "Preferred response", "rejected": "Less preferred response"}
+```
+
+if the Prompt template accept a system message, you can extend the Dataset with a additional "system" field.
+
+```jsonl
+{"system": "You are a helpfull assistant", "prompt": "User prompt", "chosen": "Preferred response", "rejected": "Less preferred response"}
+```
+
+---
+
+### GRPO-Training
+#### Overview
+
+Group Relative Policy Optimization (GRPO) is a fine-tuning method that optimizes language models by generating multiple responses per prompt and learning from their relative quality. This approach helps improve response quality through comparative learning.
+
+#### Dataset Format
+
+GRPO requires a dataset in JSONL format (one JSON object per line) with the following structure:
+
+```json
+{"prompt": "Your question or instruction here", "answer": "The expected response"}
+```
+
+Each entry must contain:
+- `prompt`: The input text for the model to respond to
+- `answer`: The target/reference response
+
+Optional fields:
+- `system`: A system message providing context or instructions for the model
+
+Example entries:
+```json
+{"prompt": "Gerald spends $100 a month on baseball supplies. His season is 4 months long. He wants to use the months he's not playing baseball to save up by raking, shoveling, and mowing lawns. He charges $10 for each. How many chores does he need to average a month to save up for his supplies?", "answer": "5"}
+{"prompt": "Ann is cutting fabric to make curtains. She cuts a 4 foot by 6 foot rectangle for the living room, and a 2 foot by 4 foot rectangle for the bedroom. If the bolt of fabric is 16 feet by 12 feet, how much fabric is left in square feet?", "answer": "160"}
+{"prompt": "Arnel had ten boxes of pencils with the same number of pencils in each box. He kept ten pencils and shared the remaining pencils equally with his five friends. If his friends got eight pencils each, how many pencils are in each box?", "answer": "5", "system": "You are a helpful math tutor."}
+```
+
+#### Usage
+
+To fine-tune a model using GRPO:
+
+```shell
+mlx_lm.lora \
+    --model <path_to_model> \
+    --train \
+    --data <path_to_data> \
+    --fine-tune-type grpo \
+    --group-size 4
+```
+
+#### GRPO-Specific Arguments
+
+- `--group-size`: Number of responses generated per prompt (default: 4)
+- `--beta`: KL penalty coefficient for policy optimization (default: 0.1)
+- `--epsilon`: Small constant for numerical stability (default: 1e-4)
+- `--max-completion-length`: Maximum length of generated completions (default: 512)
+- `--reference-model-path`: Path to reference model weights. If not specified, uses the same model
+- `--temperature`: Sampling temperature for generations. Higher values increase randomness (default: 1.0)
+- `--reward-weights`: Optional list of weights for multiple reward functions. Must match number of reward functions. If not specified, all rewards weighted equally with 1.0
+- `--use-chat-template`: Whether to use the model's chat template for formatting prompts (default: False)
+- `--use-prompt`: Whether to use the prompt as part of the input for generation (default: False)
+
+#### Training Process
+
+During GRPO training, the model:
+1. Takes each prompt from the dataset
+2. Generates multiple responses (specified by `--group-size`)
+3. Evaluates these responses against the reference answer
+4. Optimizes the policy based on the relative quality of the responses
+
+#### Resource Considerations
+
+GRPO requires more compute resources than standard LoRA training since it generates multiple responses per prompt. Consider:
+- Reducing batch size
+- Using gradient checkpointing
+- Adjusting `--group-size` to balance between quality and resource usage
+
+If running into memory issues, you can also try:
+- Reducing `--max-completion-length`
+- Using a smaller model for initial experiments
+
+#### Example Command with Full Options
+
+```shell
+mlx_lm.lora \
+    --model <path_to_model> \
+    --train \
+    --data <path_to_data> \
+    --fine-tune-type grpo \
+    --group-size 4 \
+    --beta 0.1 \
+    --epsilon 1e-4 \
+    --max-completion-length 512 \
+    --reference-model-path <optional_path_to_reference_model> \
+    --temperature 1.0 \
+    --reward-weights 1.0 1.0 \
+    --use-chat-template False \
+    --use-prompt False \
+    --batch-size 4 \
+    --learning-rate 1e-5 \
+    --num-epochs 3
+```
+
+---
+
+### Evaluate
+
+To compute test set perplexity use:
+
+```shell
+mlx_lm_lora.train \
+    --model <path_to_model> \
+    --adapter-path <path_to_adapters> \
+    --data <path_to_data> \
+    --test
+```
+
+### Generate
+
+For generation use `mlx-lm` with `mlx_lm.generate`:
+
+```shell
+mlx_lm.generate \
+    --model <path_to_model> \
+    --adapter-path <path_to_adapters> \
+    --prompt "<your_model_prompt>"
+```
+
+#### Prompt Masking
+
+The default training computes a loss for every token in the sample. You can
+ignore the prompt and compute loss for just the completion by passing
+`--mask-prompt`. Note this is only supported for `chat` and `completion`
+datasets. For `chat` datasets the final message in the message list is
+considered the completion.
+
+---
+
+## Fuse
+
+You can generate a model fused with the low-rank adapters using the
+`mlx_lm_lora.fuse` command. This command also allows you to optionally:
+
+- Upload the fused model to the Hugging Face Hub.
+- Export the fused model to GGUF. Note GGUF support is limited to Mistral,
+  Mixtral, and Llama style models in fp16 precision.
+
+To see supported options run:
+
+```shell
+mlx_lm_lora.fuse --help
+```
+
+To generate the fused model run:
+
+```shell
+mlx_lm_lora.fuse --model <path_to_model>
+```
+
+This will by default load the adapters from `adapters/`, and save the fused
+model in the path `fused_model/`. All of these are configurable.
+
+To upload a fused model, supply the `--upload-repo` and `--hf-path` arguments
+to `mlx_lm_lora.fuse`. The latter is the repo name of the original model, which is
+useful for the sake of attribution and model versioning.
+
+For example, to fuse and upload a model derived from Mistral-7B-v0.1, run:
+
+```shell
+mlx_lm_lora.fuse \
+    --model mistralai/Mistral-7B-v0.1 \
+    --upload-repo mlx-community/my-lora-mistral-7b \
+    --hf-path mistralai/Mistral-7B-v0.1
+```
+
+To export a fused model to GGUF, run:
+
+```shell
+mlx_lm_lora.fuse \
+    --model mistralai/Mistral-7B-v0.1 \
+    --export-gguf
+```
+
+This will save the GGUF model in `fused_model/ggml-model-f16.gguf`. You
+can specify the file name with `--gguf-path`.
+
+---
+
+## Memory Issues
+
+Fine-tuning a large model with LoRA requires a machine with a decent amount
+of memory. Here are some tips to reduce memory use should you need to do so:
+
+1. Try quantization (QLoRA). You can use QLoRA by generating a quantized model
+   with `convert.py` and the `-q` flag. See the [Setup](#setup) section for
+   more details.
+
+2. Try using a smaller batch size with `--batch-size`. The default is `4` so
+   setting this to `2` or `1` will reduce memory consumption. This may slow
+   things down a little, but will also reduce the memory use.
+
+3. Reduce the number of layers to fine-tune with `--num-layers`. The default
+   is `16`, so you can try `8` or `4`. This reduces the amount of memory
+   needed for back propagation. It may also reduce the quality of the
+   fine-tuned model if you are fine-tuning with a lot of data.
+
+4. Longer examples require more memory. If it makes sense for your data, one thing
+   you can do is break your examples into smaller
+   sequences when making the `{train, valid, test}.jsonl` files.
+
+5. Gradient checkpointing lets you trade-off memory use (less) for computation
+   (more) by recomputing instead of storing intermediate values needed by the
+   backward pass. You can use gradient checkpointing by passing the
+   `--grad-checkpoint` flag. Gradient checkpointing will be more helpful for
+   larger batch sizes or sequence lengths with smaller or quantized models.