Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Use this as the implementation source: clone the repo and follow its README for install, dependencies, and how to run code or experiments. The generated client prints JSON with a suggested git clone command.
This is the paper or artifact home from DOI/registry metadata — not a JSON API. If this URL is arXiv, the generated client can still fetch live Atom metadata (title, abstract, authors) without a BASE_URL. For other hosts, the client uses stub mode until you set a real BASE_URL for a REST service.
What “running” this client does
The *_client.py script prints JSON that combines a GitHub repository (clone URL + suggested git clone) with optional paper context from arXiv (live Atom metadata when reference_url is arXiv). Run the real code by cloning the repo and following its README — the skill is your agent-facing entrypoint, not a substitute for the repo’s install steps.
To call a REST API instead, set BASE_URL in scripts/text_to_lora_client.py or wrap the upstream CLI with subprocess after clone.
How to run the method (from the source)
Extracted for operators and agents. Confirm against the upstream repository or paper before relying on it in production.
Prerequisites
Python 3.10
16GB GPU memory (to run demo with both models simultaneously)
uv run huggingface-cli login
uv run huggingface-cli download SakanaAI/text-to-lora --local-dir . --include "trained_t2l/*"
Generate LoRA from task description
uv run python scripts/generate_lora.py trained_t2l/llama_8b_t2l "This task challenges your problem-solving abilities through mathematical reasoning. You must carefully read each scenario and systematically work through the data to compute the final outcome."
For smaller GPUs, use Gemma-2-2B:
uv run python scripts/generate_lora.py trained_t2l/gemma_2b_t2l "This task challenges your problem-solving abilities through mathematical reasoning. You must carefully read each scenario and systematically work through the data to compute the final outcome."
Hugging Face Auth: Run uv run huggingface-cli login before downloading models/datasets
WANDB_MODE: Set WANDB_MODE=disabled to disable Weights & Biases logging
Flash Attention wheel: The provided wheel is for CUDA 12.3 and torch 2.3. Adjust URL if your hardware differs.
vLLM non-determinism: Known issue—evaluation runs may show small variance even with fixed seed due to vLLM's LoRA implementation
Dataset connection issues: If Hugging Face datasets server rejects connections during SFT training, retry until datasets are cached locally
The same text lives inscripts/USAGE.mdfor tools that prefer reading files underscripts/.
Parameters
--api-key (str) [required] API key for authentication
--t2l-directory (str) [required] Path to the trained T2L model directory (e.g., trained_t2l/llama_8b_t2l)
--task-description (str) [required] Natural language description of the task for which to generate a LoRA adapter
--model-dir (str) [optional, default=None] Base model directory for evaluation (e.g., meta-llama/Llama-3.1-8B-Instruct)
--lora-dirs (str) [optional, default=None] Path to generated LoRA directory for evaluation
--tasks (str) [optional, default=None] Comma-separated list of evaluation tasks (e.g., gsm8k)
--save-results (bool) [optional, default=False] Save evaluation results to disk
--use-icl (bool) [optional, default=False] Include 3-shot in-context examples in evaluation queries
--checkpoint-path (str) [optional, default=None] Path to T2L hypermod checkpoint for evaluation
Usage
python3 scripts/text_to_lora_client.py uv run python scripts/generate_lora.py trained_t2l/llama_8b_t2l "This task challenges your problem-solving abilities through mathematical reasoning."
