Reuse
Borrowing it
Nothing to install: this file belongs to ProfSynapse/Synaptic-Tuner. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/ProfSynapse/Synaptic-Tuner/main/.agents/skills/fine-tuning/SKILL.mdgit clone --depth 1 https://github.com/ProfSynapse/Synaptic-TunerWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/profsynapse/synaptic-tuner/fine-tuning)<a href="https://agentmods.dev/skills/profsynapse/synaptic-tuner/fine-tuning"><img src="https://agentmods.dev/badge/skills/profsynapse/synaptic-tuner/fine-tuning/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/profsynapse/synaptic-tuner/fine-tuning"><img src="https://agentmods.dev/badge/skills/profsynapse/synaptic-tuner/fine-tuning.svg" alt="Reviewed on agentmods" width="80" height="20"></a> Per session 113 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 12,453 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits How audits are shown
- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium MCP Rug Pull · line 139 Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.Fix: Pin the image: image:tag or image@sha256:abc123
- medium Agent Snooping · line 147 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
Origin original No closer match found in the catalogue.
Token cost
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00113 | $0.12453 |
| Opus 5.5 | $0.00045 | $0.04981 |
| Sonnet 5.5 | $0.00023 | $0.02491 |
| Haiku 4.5 | $0.00011 | $0.01245 |
Security
Grade A, and why
fine-tuning scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
.agents/skills/fine-tuning/SKILL.md · 686 lines
How it starts
The opening of the file, as written. The whole thing — 686 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fine-Tuning Pipeline
Train language models with SFT, KTO, and GRPO locally or on supported cloud providers. This skill also covers the Karpathy-style experiment loop, checkpoint evaluation, LoRA surgery, and the current HF Jobs operational path.
Quick Reference
| Task | Command |
|---|---|
| Interactive menu | ./run.sh → Train |
| Local Docker config run | python tuner.py local-run --job-config Trainers/recipes/<recipe>.yaml --yes |
| SFT training | cd Trainers/sft && python train_sft.py --model-size 7b |
| KTO training | cd Trainers/kto && python train_kto.py --model-size 7b |
| Modal conversation SFT host path | Existing train --job-config <recipe> --plan --json, explicit conversation format and compatible named candidate; CPU host proof, unqualified runtime; see reference/conversation-preparation.md |
| Public GRPO preparation and Modal runner | Explicit host-owned preparation ports and compatible named GRPO profile; bounded live Modal mechanics evidence is source-scoped, with broader ordinary runtime/lifecycle gates retained; see reference/grpo-training.md and the support matrix |
| Legacy GRPO training | cd Trainers/grpo && python train_grpo.py |
| Pivot-profile GRPO dataset | cd Trainers/grpo && python train_grpo.py --config configs/pivot_config.yaml --pivot-profile-only |
| GRPO with pivot filtering | cd Trainers/grpo && python train_grpo.py --config configs/pivot_config.yaml |
| Env-backed GRPO | cd Trainers/grpo && python train_env_grpo.py --config ./configs/env_config.yaml --dry-run |
| Experiment loop | python tuner.py experiment-loop --experiment-config configs/flywheel/experiment_loop.yaml |
| LoRA surgery | python tuner.py surgery --surgery-config configs/lora_surgery.yaml |
| HF custom job | python tuner.py cloud-run --job-config Trainers/recipes/<recipe>.yaml |
| Canonical HF train+eval | python tuner.py cloud-pipeline --method sft --preset full |
| Full experiment bundle | python tuner.py run-experiment --experiment-spec Trainers/cloud/experiments/<spec>.yaml --yes |
| Evolutionary SFT smoke test | python tuner.py run-experiment --experiment-spec Trainers/cloud/experiments/<evolutionary-spec>.yaml --yes |
| Staggered experiment batch | python3 scripts/launch_experiment_batch.py Trainers/cloud/experiments/<spec1>.yaml Trainers/cloud/experiments/<spec2>.yaml --yes |
| One-shot RunPod wrapper job | python3 scripts/runpod_run_job.py --run-tag <tag> --repo-url <git-url> --commit <full-sha> --wrapper <repo-relative.sh> --dry-run |
| Modal training | Public TrainingAPI with a host-owned durable grant; no manual modal run path |
| Modal CPU progress-transport proof | Recipe observation_capture plus python tuner.py train --job-config <reviewed-modal-recipe.yaml> --qualify --modal-profile <name> --modal-environment <existing> --json; same normal CPU self-check, no GPU/retraining; see reference/modal-jobs.md |
| Modal local input preparation | python tuner.py train --job-config <modal-recipe.yaml> --prepare-local --json — optional pinned public downloads/source-wheel preflight; normal execution performs the same preparation before claims/effects; see reference/modal-jobs.md |
| Modal release-lookup diagnosis | python scripts/inspect_modal_release_lookup.py --journal <exact-private-journal> --claim-ref deploy-<digest> --environment <name> --modal-profile <profile> — read-only, non-authorizing; see reference/modal-jobs.md |
| Modal rate observation | python -B examples/modal_chat/launch.py --project-root <consumer> --configuration <config> --mode quote-training --modal-profile <name> — read-only; see reference/modal-jobs.md |
| Chat with a verified SFT run | Embedded open_run_chat with a consumer-owned RunsAPI and runtime adapter; see docs/architecture/verified-run-chat.md (repo root) |
| Blind hardware plan | python tuner.py plan-hardware --experiment-spec Trainers/cloud/experiments/<spec>.yaml |
| Analyze finished experiment | python tuner.py analyze-experiment --experiment-id latest |
| Analyze/prune dataset from loss | python3 scripts/prune_dataset_from_loss.py --dataset-path ... --experiment-id ... --analyze-only |
| Train/eval contamination check | python tuner.py check-contamination --train-data Datasets/<train>.jsonl |
| Grouped validation split | recipe dataset: {split_dataset: true, validation_group_key: metadata.scenario} |
| Standalone prompt optimization | python tuner.py prompt-optimize --prompt-opt-config configs/prompt_optimization/NAME.yaml |
| Prompt-optimize SynthChat generation | python -m SynthChat.run generate --prompt-opt-config configs/prompt_optimization/NAME.yaml [options] |
| Analyze bucket-backed run | python tuner.py bucket analyze --path runs/hf_jobs/sft/<run-prefix>/ |
| Read bucket artifact | python tuner.py bucket read --path runs/.../logs/training_latest.jsonl --jsonl-latest --pretty |
| List bucket prefix | python tuner.py bucket list --path runs/hf_jobs/sft/<run-prefix>/ --limit 20 |
| Pull bucket prefix locally | python tuner.py bucket pull --path runs/hf_jobs/sft/<run-prefix>/ --dest . |
| Push local artifact to bucket | python tuner.py bucket push --path local/results.json --dest runs/manual_uploads/ |
| Live HF job list | python tuner.py cloud-jobs list |
| Live HF job logs | python tuner.py cloud-jobs logs --job professorsynapse/<job-id> --tail 200 |
| Cloud eval against a run | python tuner.py cloud-eval --run latest --preset full |
| Local deterministic vLLM generation | VLLM_BATCH_INVARIANT=1 python tuner.py batch-generate --engine vllm ... |
| HF gym against trained model | python tuner.py cloud-gym --run latest --method sft |
| Prepare verified raw-text SFT data | python tuner.py prepare-dataset --config <config.json> --json |
| Check SFT loss masking (tokenizer only) | python tuner.py doctor sft-mask --dataset-path <jsonl> --model <tokenizer> |
| Inspect native training host candidate | python tuner.py doctor native-training --json — read-only platform/Python/package metadata; see reference/native-local-training.md |
| Prepare representative native SFT subsets | Config-first examples/training_qualification/native_sft_datasets/prepare.py {acquire,prepare}; existing conversation publisher, explicit labels/splits, no training authority; see its README |
| Diagnose an offline SFT worker rejection | Read the exact retained trainer.stderr.log; closed phase/category/milestone labels locate a boundary, not a cause or retry authority; see reference/sft-training.md |
| Diagnose Torch import in a retained Docker image | python scripts/diagnose_docker_torch_import.py --help for bounded guarded, worker-context prefix, and worker-entry probes; each live command needs exact review; see reference/sft-training.md |
| Native MLX restricted host consumer | python -B examples/training_qualification/native_host.py inspect <canonical-host-config.json> is read-only; v1 start or v2 lifecycle qualify requires an exact externally signed single-use admission and separate public-key pin. Bounded M4 SFT/GRPO and public lifecycle qualification passed; every new attempt still requires reviewed source/runtime and fresh authority. See reference/native-local-training.md |
| Native MLX GRPO | Same public native composition and restricted host consumer, separate named profile/54-package runtime closure, attributed on-policy objective and bounded rewards. Bounded M4 mechanics and lifecycle evidence passed; ordinary qualification stays host-authenticated. See reference/native-local-training.md and reference/grpo-training.md |
| Diagnose a retained native GRPO failure | Read the exact authenticated process outcome with its admitted source. Closed numeric exits locate phases only; historical 122 remains unclassified. See reference/native-local-training.md; no replay or raw exception/traceback logging. |
| Native-only GRPO input preparation | python -B examples/training_qualification/native_grpo_preparation/prepare.py {inspect,prepare} <reviewed-config.json> — existing generic preparation service through a config-first Linux host wrapper, no Modal/runtime/ML effects; read the example README and reference/grpo-training.md |
| Native GRPO tokenizer-only profile | python -B examples/training_qualification/native_grpo_preparation/prepare.py profile-tokenizer <reviewed-tokenizer-profile.json> — Linux/Darwin host, exact shared prompt-ID contract and private bounded evidence; root reviews source/environment first, no MLX/model/training; see the example README. |
| Capture reviewed native runtime bytes | python -B scripts/capture_native_mlx_runtime.py --config <absolute-config> --dependency-lock <absolute-lock> --output <new-private-file> — observation only; explicit roots, dependencies and original pins required; see reference/native-local-training.md |
| Plan private native Mac runtime | python -B examples/training_qualification/native_runtime_preparation.py plan <absolute-reviewed-config> — local 54-wheel archive plan only; later acquire/install consume separate private claims. seal-uv-locks is a separately reviewed one-use custody repair for an exact completed install; see reference/native-local-training.md |
| Prepare a reviewed private native model snapshot | python -B examples/training_qualification/native_model_preparation.py inspect <absolute-config> is local-only; metadata-candidate performs one bounded SDK metadata request and acquire downloads only after separate exact inventory review. Neither grants training admission; see reference/native-local-training.md |
| Plan a derived training image | python scripts/qualify_derived_training_image.py plan --config Trainers/image_profiles/<profile>.yaml |
| Warm Space scaffold | python3 Trainers/cloud/scripts/manage_space.py render --template vllm_warm --output-dir /tmp/my-space --base-image ghcr.io/<org>/<image>:<tag> |
| Warm Space deploy | python3 Trainers/cloud/scripts/manage_space.py deploy --space-id <user>/<space> --template vllm_warm --base-image ghcr.io/<org>/<image>:<tag> --hardware a10g-small --sleep-time 3600 --var BASE_MODEL=<model> |
| ML training | python tuner.py ml train --config Trainers/ml/configs/templates/regression.yaml |
| Decision model (Jev-style) smoke | python tuner.py local-run --job-config Trainers/recipes/decision_qwen35_2b_pointer_smoke.yaml --yes |
| Image LoRA (Qwen-Image, Modal) | python Trainers/image_lora/train_image_lora.py build-dataset --recipe <private recipe> --out <dataset>, then plan --dataset <dataset>, probe, launch --dataset <dataset> --output-dir <dir> --max-usd <cap>, status/fetch/cleanup --state <dir>/run_state.json; see reference/image-lora.md |
Files
What ships with it
48 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- configs/dora.yaml 971 B
- configs/eva.yaml 1.1 KB
- configs/experiment_model_gpu_template.yaml 1.9 KB
- configs/gemma4_e4b_a100.yaml 1.3 KB
- configs/grpo_minimal.yaml 813 B
- configs/loftq.yaml 777 B
- configs/nexus_note_snapshot.example.yaml 463 B
- configs/olora.yaml 851 B
- configs/pissa.yaml 932 B
- configs/qlora_dora.yaml 802 B
- configs/qwen3_4b_a100.yaml 1.0 KB
- configs/qwen35_4b_a100.yaml 1.3 KB
- configs/qwen35_4b_token_profile.yaml 1.1 KB
- configs/regret_free.yaml 1.3 KB
- protocols/environment-backed-alignment-pipeline.md 10 KB
- reference/certificate-trust.md 5.3 KB
- reference/checkpoint-evaluation.md 2.2 KB
- reference/cloud-experiment-launching.md 7.2 KB
- reference/cloud-training.md 51 KB
- reference/conversation-preparation.md 13 KB
- reference/dataset-formats.md 20 KB
- reference/decision-training.md 9.1 KB
- reference/derived-training-images.md 5.3 KB
- reference/experiment-loop.md 4.3 KB
- reference/grpo-training.md 37 KB
- reference/hf-spaces-warm-iteration.md 2.6 KB
- reference/image-lora.md 4.3 KB
- reference/kto-training.md 5.4 KB
- reference/lora-surgery.md 4.8 KB
- reference/lora-techniques.md 12 KB
- reference/modal-jobs.md 126 KB
- reference/model-presets.md 5.1 KB
- reference/native-local-training.md 44 KB
- reference/nexus-note-snapshot.md 1.7 KB
- reference/runpod-jobs.md 13 KB
- reference/sft-training.md 35 KB
- reference/tokenizer-profiling.md 5.5 KB
- reference/training-config.md 16 KB
- reference/troubleshooting.md 5.4 KB
- scripts/hf_jobs_hardware.py 4.9 KB runs code
- scripts/launch_experiment_batch.py 4.0 KB runs code
- scripts/profile_tokenizer_lengths.py 64 KB runs code
- scripts/prune_dataset_from_loss.py 25 KB runs code
- scripts/read_bucket_artifact.py 1.2 KB runs code
- scripts/runpod_run_job.py 16 KB runs code
- scripts/snapshot_nexus_notes.py 9.7 KB runs code
- scripts/split_for_gspo.py 5.9 KB runs code
- skill-spec.md 4.3 KB
Changes
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago Changed a502e4de431d
- 9d ago Changed · +5 lines 6fa0d973e0ba
- 10d ago Changed · +6 lines 378b539bdbc0
- 14d ago Changed · +6 lines 5b3f9a798099
- 17d ago Changed · +1 lines 3f3f65dc7e5f
- 23d ago Changed 20b8d8390d6d
- 24d ago Changed · +2 lines d6df017f28a8
- 25d ago Changed · +2 lines 25ac9a608279
- 27d ago Changed · +4 lines a2bebf7d1970
- 1mo ago First seen · 660 lines · 113 tokens per session scan A f0c003c49ff1
fine-tuning is a skill published in the GitHub repository ProfSynapse/Synaptic-Tuner (28 stars, last pushed 3d ago), licensed MIT. It adds 113 tokens to every session and 12,453 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Related
fine-tuning compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| fine-tuningProfSynapse/Synaptic-Tuner | Skill | 28 | 3d ago | MIT | A | 113 | Claude Code |
| arboretoK-Dense-AI/scientific-agent-skills | Skill | 56k | 4d ago | MIT | A | 55 | Claude Code, Codex |
| deepspot-mK-Dense-AI/scientific-agent-skills | Skill | 56k | 4d ago | MIT | A | 70 | Claude Code |
| torchdrugK-Dense-AI/scientific-agent-skills | Skill | 56k | 4d ago | MIT | A | 64 | Claude Code |
| pyhealthK-Dense-AI/scientific-agent-skills | Skill | 56k | 4d ago | MIT | A | 84 | Claude Code, Codex |
| nemo-mbridge-perf-expert-parallel-overlapNVIDIA/skills | Skill | 3.9k | yesterday | Apache-2.0 | A | 56 | Claude Code |
| ml-training-recipesOrchestra-Research/AI-Research-SKILLs | Skill | 14k | 27d ago | MIT | A | 88 | Claude Code |