ml-training plugin
Model architectures, tokenization, fine-tuning, post-training, distributed training, optimization.
How to install the ml-training plugin
Run both commands inside Claude Code. The first adds the marketplace once; the second installs the plugin.
/plugin marketplace add KalarisLabs/research-agent-skills
/plugin install ml-training@research-agent-skillsWhat's inside ml-training
32 parts read from the plugin's own files.
Skills 32
Claude loads it when the task matches- awq-quantization
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.
- axolotl
Provides guidance for fine-tuning large language models with Axolotl, covering YAML training configs, LoRA and QLoRA, preference training with DPO, KTO, ORPO and GRPO, multimodal models, FSDP and multi-GPU setups, sequence (context) parallelism, NCCL bandwidth tests, dataset formats, compressed model saving for vLLM and llmcompressor, and custom integrations. Use when writing or debugging an Axolotl YAML config, choosing a dataset format for a fine-tuning run, setting up FSDP or context_parallel_size across GPUs, or running DPO, KTO, ORPO or GRPO training. Use when saving a compressed model for vLLM inference, or when writing a custom Axolotl plugin or integration. Not for inference serving or general Hugging Face Trainer scripts outside Axolotl.
- deepspeed
Covers DeepSpeed for distributed deep learning training and I/O: ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8 training, 1-bit Adam, sparse attention, and DeepNVMe (aio_handle, gds_handle, async_io and gds operators, ds_nvme_tune, ds_report) for fast transfers between NVMe storage and host or GPU tensors. Use when configuring ZeRO stages for large-model training, enabling mixed precision or 1-bit Adam, offloading parameters or optimizer state to NVMe with ZeRO-Infinity, writing or reading tensors to files with blocking or non-blocking DeepNVMe calls, or tuning NVMe I/O settings such as block size, queue depth and parallelism. Not for plain PyTorch DistributedDataParallel without DeepSpeed.
- distributed-llm-pretraining-torchtitan
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
- fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
- gguf-quantization
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.
- gptq
Quantizes LLMs to 4-bit (also 3-bit) with GPTQ using group-wise quantization (group size 128 by default), via AutoGPTQ and transformers. Covers loading pre-quantized GPTQ models, quantizing your own model, choosing group size, selecting ExLlamaV2, Marlin, or Triton kernels, and QLoRA fine-tuning with PEFT. Use when fitting 70B-class models onto limited or consumer GPUs, cutting memory about 4x versus FP16, speeding up inference, finding pre-quantized checkpoints on HuggingFace, or fine-tuning a quantized model with LoRA. For slightly better accuracy on newer GPUs use AWQ instead, and for simple 8-bit or on-the-fly quantization use bitsandbytes.
- grpo-rl-training
Guides GRPO (Group Relative Policy Optimization) fine-tuning of language models with the TRL library, including GRPOTrainer configuration, composing multiple reward functions (correctness, format, length, style), dataset prep in chat format, Unsloth setup, LoRA merging, and monitoring reward, reward_std and KL. Use when training a model to follow a strict output format such as XML or JSON. Use when teaching verifiable tasks like math or code with objective correctness rewards. Use when improving chain-of-thought reasoning with custom reward functions. Use when debugging flat rewards, mode collapse, or OOM in GRPO runs. Do not use for plain supervised fine-tuning (use SFT) or when high-quality preference pairs exist (use DPO or PPO).
- hqq-quantization
Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.
- huggingface-accelerate
Wraps existing PyTorch training scripts with HuggingFace Accelerate (Accelerator class, accelerate config, accelerate launch) so the same code runs on CPU, single GPU, multi-GPU, multi-node, TPU, or Apple MPS. Covers device placement, FP16/BF16/FP8 mixed precision, gradient accumulation, distributed checkpointing, and switching between DDP, DeepSpeed ZeRO, FSDP, and Megatron backends. Use when converting a single-GPU script to multi-GPU, enabling mixed precision, configuring DeepSpeed ZeRO or FSDP, or setting up gradient accumulation. Use when one script must run on different hardware. Not for callback-heavy training loops (use PyTorch Lightning) or multi-node orchestration with hyperparameter tuning (use Ray Train).
- huggingface-tokenizers
Provides the HuggingFace Tokenizers library (Rust core with Python and Node.js bindings) for training and using BPE, WordPiece, and Unigram tokenizers. Covers the pipeline of normalizers, pre-tokenizers, models, post-processors and decoders, padding and truncation, batch encoding, alignment tracking, and conversion to transformers PreTrainedTokenizerFast. Use when training a custom tokenizer or vocabulary on a new corpus, tokenizing large text corpora quickly, mapping tokens back to character offsets for NER or question answering, or configuring normalization and special tokens. Use when wrapping a custom tokenizer for transformers. For SentencePiece models or tiktoken, use those tools instead; for loading a pretrained tokenizer only, AutoTokenizer is enough.
- implementing-llms-litgpt
No description in the file.
- llama-factory
No description in the file.
- mamba-architecture
No description in the file.
- miles-rl-training
No description in the file.
- ml-training-recipes
No description in the file.
- nanogpt
No description in the file.
- openrlhf-training
No description in the file.
- optimizing-attention-flash
No description in the file.
- peft-fine-tuning
No description in the file.
- pytorch-fsdp2
No description in the file.
- pytorch-lightning-distributed
No description in the file.
- quantizing-models-bitsandbytes
No description in the file.
- ray-train
No description in the file.
- rwkv-architecture
No description in the file.
- sentencepiece
No description in the file.
- simpo-training
No description in the file.
- slime-rl-training
No description in the file.
- torchforge-rl-training
No description in the file.
- training-llms-megatron
No description in the file.
- unsloth
No description in the file.
- verl-rl-training
No description in the file.
Using ml-training in Claude Code
Skills need no command. Claude reads them when your request matches what they describe.
Update, turn off or remove
- Update the plugin (shell)
- claude plugin update ml-training@research-agent-skills
- Refresh the marketplace
- /plugin marketplace update research-agent-skills
- Turn it off for now
- /plugin disable ml-training@research-agent-skills
- Remove it
- /plugin uninstall ml-training@research-agent-skills
Skills in ml-training
- awq-quantizationKalarisLabs/research-agent-skills
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.
3 files126stars - axolotlKalarisLabs/research-agent-skills
Provides guidance for fine-tuning large language models with Axolotl, covering YAML training configs, LoRA and QLoRA, preference training with DPO, KTO, ORPO and GRPO, multimodal models, FSDP and multi-GPU setups, sequence (context) parallelism, NCCL bandwidth tests, dataset formats, compressed model saving for vLLM and llmcompressor, and custom integrations. Use when writing or debugging an Axolotl YAML config, choosing a dataset format for a fine-tuning run, setting up FSDP or context_parallel_size across GPUs, or running DPO, KTO, ORPO or GRPO training. Use when saving a compressed model for vLLM inference, or when writing a custom Axolotl plugin or integration. Not for inference serving or general Hugging Face Trainer scripts outside Axolotl.
5 files126stars - deepspeedKalarisLabs/research-agent-skills
Covers DeepSpeed for distributed deep learning training and I/O: ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8 training, 1-bit Adam, sparse attention, and DeepNVMe (aio_handle, gds_handle, async_io and gds operators, ds_nvme_tune, ds_report) for fast transfers between NVMe storage and host or GPU tensors. Use when configuring ZeRO stages for large-model training, enabling mixed precision or 1-bit Adam, offloading parameters or optimizer state to NVMe with ZeRO-Infinity, writing or reading tensors to files with blocking or non-blocking DeepNVMe calls, or tuning NVMe I/O settings such as block size, queue depth and parallelism. Not for plain PyTorch DistributedDataParallel without DeepSpeed.
10 files126stars - distributed-llm-pretraining-torchtitanKalarisLabs/research-agent-skills
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
5 files126stars - fine-tuning-with-trlKalarisLabs/research-agent-skills
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
5 files126stars - gguf-quantizationKalarisLabs/research-agent-skills
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.
3 files126stars - gptqKalarisLabs/research-agent-skills
Quantizes LLMs to 4-bit (also 3-bit) with GPTQ using group-wise quantization (group size 128 by default), via AutoGPTQ and transformers. Covers loading pre-quantized GPTQ models, quantizing your own model, choosing group size, selecting ExLlamaV2, Marlin, or Triton kernels, and QLoRA fine-tuning with PEFT. Use when fitting 70B-class models onto limited or consumer GPUs, cutting memory about 4x versus FP16, speeding up inference, finding pre-quantized checkpoints on HuggingFace, or fine-tuning a quantized model with LoRA. For slightly better accuracy on newer GPUs use AWQ instead, and for simple 8-bit or on-the-fly quantization use bitsandbytes.
4 files126stars - grpo-rl-trainingKalarisLabs/research-agent-skills
Guides GRPO (Group Relative Policy Optimization) fine-tuning of language models with the TRL library, including GRPOTrainer configuration, composing multiple reward functions (correctness, format, length, style), dataset prep in chat format, Unsloth setup, LoRA merging, and monitoring reward, reward_std and KL. Use when training a model to follow a strict output format such as XML or JSON. Use when teaching verifiable tasks like math or code with objective correctness rewards. Use when improving chain-of-thought reasoning with custom reward functions. Use when debugging flat rewards, mode collapse, or OOM in GRPO runs. Do not use for plain supervised fine-tuning (use SFT) or when high-quality preference pairs exist (use DPO or PPO).
5 files126stars - hqq-quantizationKalarisLabs/research-agent-skills
Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.
3 files126stars - huggingface-accelerateKalarisLabs/research-agent-skills
Wraps existing PyTorch training scripts with HuggingFace Accelerate (Accelerator class, accelerate config, accelerate launch) so the same code runs on CPU, single GPU, multi-GPU, multi-node, TPU, or Apple MPS. Covers device placement, FP16/BF16/FP8 mixed precision, gradient accumulation, distributed checkpointing, and switching between DDP, DeepSpeed ZeRO, FSDP, and Megatron backends. Use when converting a single-GPU script to multi-GPU, enabling mixed precision, configuring DeepSpeed ZeRO or FSDP, or setting up gradient accumulation. Use when one script must run on different hardware. Not for callback-heavy training loops (use PyTorch Lightning) or multi-node orchestration with hyperparameter tuning (use Ray Train).
4 files126stars - huggingface-tokenizersKalarisLabs/research-agent-skills
Provides the HuggingFace Tokenizers library (Rust core with Python and Node.js bindings) for training and using BPE, WordPiece, and Unigram tokenizers. Covers the pipeline of normalizers, pre-tokenizers, models, post-processors and decoders, padding and truncation, batch encoding, alignment tracking, and conversion to transformers PreTrainedTokenizerFast. Use when training a custom tokenizer or vocabulary on a new corpus, tokenizing large text corpora quickly, mapping tokens back to character offsets for NER or question answering, or configuring normalization and special tokens. Use when wrapping a custom tokenizer for transformers. For SentencePiece models or tiktoken, use those tools instead; for loading a pretrained tokenizer only, AutoTokenizer is enough.
6 files126stars
ml-training plugin questions
How do I install the ml-training plugin?
In Claude Code run /plugin marketplace add KalarisLabs/research-agent-skills, then /plugin install ml-training@research-agent-skills.
What does the ml-training plugin include?
ml-training bundles 32 skills.
How do I update or remove the ml-training plugin?
Update it with claude plugin update ml-training@research-agent-skills in your shell, or from the Installed tab of /plugin. Remove it with /plugin uninstall ml-training@research-agent-skills.
What license is the ml-training plugin under?
MIT, as stated in the plugin's manifest.