skill CX Codex

bailian-finetune — Claude Code skill: what it does, how to install

Q: How do I install bailian-finetune in Claude Code?

Run `npx skills add modelstudioai/cli --skill bailian-finetune -a claude-code`. The skill file lands at `~/.claude/skills/bailian-finetune/SKILL.md`.

Q: What does bailian-finetune do?

It runs the Bailian fine-tuning loop through `bl`: dataset validation and upload, job creation, progress and logs, checkpoint selection, export, optional compression, then deployment of the resulting model as a service.

Q: How does it differ from the other Bailian skills?

Picking a model belongs to `bailian-model-recommend`, image and video generation with ready models to `bailian-gen`, and general resource management to `bailian-cli`. Fine-tuning on a platform the user names explicitly is not handled here.

Drives a Bailian fine-tuning job end to end, from dataset check to a deployed service, with sft, lora, dpo and cpt on text, audio and image models.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓

npx skills add modelstudioai/cli --skill bailian-finetune -a claude-code

Open on GitHub

What bailian-finetune does

The skill walks the whole pipeline in order: validate the dataset, upload it and get a file-id, create the job, watch the run, pick a checkpoint, export the model, compress it if you want, deploy a service. Training types are sft, sft-lora, dpo, dpo-lora and cpt. Text, audio and image bases all go through it, and bl finetune capability tells you which methods a given base accepts.

Mostly it is a dispatcher. Command detail sits in the reference/ files for dataset, finetune and deploy, and flags come from bl <command> --help rather than from the skill text, so drift against the real CLI shows up fast.

Have the bl binary and an API key ready first. The shared bailian-protocol file has to be there too: bl skill init installs it alongside the skill, and the skill halts when that file is missing. For high-risk writes such as capacity create, scale, renew, delete and overflow it advises a --dry-run preview, and --yes is injected by the runtime, so do not type it yourself. Read the result from operation_status instead of trusting HTTP 200, and remember that deploy list --status filters only the page already fetched while the server total stays unfiltered.

What it helps with

check a .jsonl against the required schema before upload
upload a dataset and keep the returned file-id
create an sft-lora job on a chosen base model
pick a checkpoint and export the model from it
deploy a service from the tuned model and read its state

Example prompts for the agent

Validate train.jsonl as chatml and upload it as my fine-tuning dataset
Create an sft-lora job on a base model from bl finetune capability and follow it with watch
Export checkpoint ckpt-N from job ft-xxx as a model named my-model

How to install bailian-finetune

1Skills CLI (Vercel)

npx skills add modelstudioai/cli --skill bailian-finetune -a claude-code

2Manually: put the files in place

~/.claude/skills/bailian-finetune/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/bailian-finetune/

1Skills CLI (Vercel)

npx skills add modelstudioai/cli --skill bailian-finetune -a codex

2Manually: put the files in place

.agents/skills/bailian-finetune/SKILL.md

For all projects use ~/.agents/skills/

1Skills CLI (Vercel)

npx skills add modelstudioai/cli --skill bailian-finetune -a cursor

2Manually: put the files in place

.cursor/skills/bailian-finetune/SKILL.md

Cursor also reads .agents/skills/ and .claude/skills/

1GitHub CLI

gh skill install modelstudioai/cli bailian-finetune

2Manually: put the files in place

.github/skills/bailian-finetune/SKILL.md

1Skills CLI (Vercel)

npx skills add modelstudioai/cli --skill bailian-finetune -a gemini-cli

2Platform CLI

gemini skills install https://github.com/modelstudioai/cli --path skills/bailian-finetune

Manually: put the files in place

.devin/skills/bailian-finetune/SKILL.md

The legacy .windsurf/skills/ path still works

Skills CLI (Vercel)

npx skills add modelstudioai/cli --skill bailian-finetune -a opencode

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about bailian-finetune

How do I install bailian-finetune in Claude Code?

Run npx skills add modelstudioai/cli --skill bailian-finetune -a claude-code. The skill file lands at ~/.claude/skills/bailian-finetune/SKILL.md.

What does bailian-finetune do?

It runs the Bailian fine-tuning loop through bl: dataset validation and upload, job creation, progress and logs, checkpoint selection, export, optional compression, then deployment of the resulting model as a service.

How does it differ from the other Bailian skills?

Picking a model belongs to bailian-model-recommend, image and video generation with ready models to bailian-gen, and general resource management to bailian-cli. Fine-tuning on a platform the user names explicitly is not handled here.

Source text

SKILL.md · Apache-2.0

---
name: bailian-finetune
metadata:
  version: "2.5.1"
  requires:
    bins: ["bl"]
description: >-
  阿里云百炼模型精调训练入口：用户要精调、微调、训练自己的模型（fine-tune，支持 SFT / SFT-LoRA / DPO / DPO-LoRA / CPT，
  覆盖文本、语音、图像）、校验或上传训练数据集、看训练进度和日志、挑 checkpoint、导出精调产物、
  把专属模型部署成服务时使用 `bl dataset` / `bl finetune` / `bl deploy`。链路是 validate 校验数据 →
  upload 拿 file-id → finetune create 建任务 → watch 看进度 → export 导出 → 可选 compress 量化压缩 → deploy 上线，需要 API key；
  写操作先用 `--dry-run` 预览。反触发：用户点名火山方舟/ark 的精调不走本 skill；只是要选哪个模型走
  bailian-model-recommend；用现成模型生图生视频走 bailian-gen；百炼其他资源管理走 bailian-cli。
  官方安装：`bl skill init`（与共享协议 bailian-protocol 同装）。
---

# Bailian fine-tuning pipeline (`bl dataset` / `bl finetune` / `bl deploy`)

**CRITICAL — Before executing, MUST read the shared protocol in [`../bailian-protocol/SKILL.md`](../bailian-protocol/SKILL.md): Version & updates (pre-flight checklist), Call attribution (every Agent `bl` command), Setup & auth, and CLI errors: report an issue. Command details are authoritative in [`reference/`](reference/index.md) (dataset / finetune / deploy) and `bl <command> --help` — do not guess flags. Examples in `reference/` stay bare; wrap with Call attribution when running. The whole pipeline requires an API key. If that protocol file is missing, stop and run `bl skill init`; do not guess auth/consent.**

**Call attribution:** Examples below and in `reference/` stay bare `bl …`. When running a command, wrap it with [`bailian-protocol` → Call attribution](../bailian-protocol/SKILL.md#call-attribution-mandatory-for-every-agent-bl-command) (bash or PowerShell). One sample for this skill (`BAILIAN_SKILL_NAME=bailian-finetune`; replace `codex` with this host). After deploy, `bl text chat` is owned by `bailian-cli`.

```bash
BAILIAN_CALLER_TYPE=agent BAILIAN_CALLER_NAME=codex BAILIAN_SKILL_NAME=bailian-finetune BAILIAN_SKILL_VERSION=2.5.0 bl dataset validate --file train.jsonl
```

## End-to-end workflow (follow in order)

```
1. Validate data   bl dataset validate --file train.jsonl [--schema chatml|dpo|cpt|tts|image]
2. Upload data     bl dataset upload --file train.jsonl          # returns a file-id
3. Create job      bl finetune text|audio|image create --base-model <base> --datasets <file-id|path>
4. Watch progress  bl finetune watch --job-id ft-xxx             # or get / logs
5. Pick artifact   bl finetune checkpoints --job-id ft-xxx
6. Export model    bl finetune export --job-id ft-xxx --checkpoint ckpt-N --model-name my-model
7. Compress model  bl finetune compress templates --model my-model       # optional; full-parameter models only
                   bl finetune compress create --model my-model --template-id <template-id>
                   bl finetune compress watch --job-id quant-xxx --follow
8. Deploy service  bl deploy text|audio|image create --model-name <my-model|quantized-output> --display-name my-svc
```

- Unsure which training methods a base model supports → `bl finetune capability --base-model <base>` or `--training-type sft|sft-lora|dpo|cpt`.
- Text `--training-type` values: `sft` / `sft-lora` / `dpo` / `dpo-lora` / `cpt`. Audio bases include `cosyvoice-v3-flash`; image bases include `wan2.7-image-pro`.
- Deployment plans: audio defaults to `--plan mu`; text/image default to `lora`.
- For `risk: high` or `requires_confirmation`, follow `bailian-protocol`; never add `--yes` automatically.

## When to use which command

| Intent                                 | Command                                                                                          |
| -------------------------------------- | ------------------------------------------------------------------------------------------------ |
| Validate / upload training data        | `bl dataset validate` / `upload` (`.jsonl` or `.zip`)                                            |
| Dataset list / detail / delete         | `bl dataset list` / `get` / `delete`                                                             |
| Create a fine-tuning job               | `bl finetune text\|audio\|image create`                                                          |
| Job list / detail / follow             | `bl finetune list` / `get` / `watch` / `logs`                                                    |
| Artifacts and export                   | `bl finetune checkpoints` / `export`                                                             |
| Quantize / compress a full model       | `bl finetune compress templates` / `create` / `watch` / `get` / `logs` / `list`                  |
| Cancel / delete a compression job      | `bl finetune compress cancel` / `delete`                                                         |
| Cancel / delete a fine-tuning job      | `bl finetune cancel` / `delete`                                                                  |
| Trainable capability lookup            | `bl finetune capability`                                                                         |
| Deploy / lifecycle                     | `bl deploy text\|audio\|image create`, `list` / `get` / `update` / `scale` / `delete` / `models` |
| Query throughput reservations          | `bl deploy list --plan ptu` / `bl deploy get`                                                    |
| Query capacity instances               | `bl deploy capacity list` / `get`                                                                |
| Query / wait for a capacity operation  | `bl deploy operation get` / `wait`                                                               |
| Buy / scale / renew / release capacity | `bl deploy capacity create` / `scale` / `renew` / `delete`                                       |
| Unsubscribe a prepaid instance         | `bl deploy capacity unsubscribe` (builds the billing console refund link)                        |
| Configure ModelCode overflow strategy  | `bl deploy overflow`                                                                             |

The `capacity list` / `get`, `operation get` / `wait` and `deploy list` / `get` queries are read-only. Capacity values are kTPM; effective, configured and target capacities are distinct. `deploy list --status` filters only the fetched page locally, with the server total left unfiltered. `operation get` reports status as data; `operation wait` exits non-zero on failure or timeout and refreshes capacity after success. Use IDs returned by the API; waiting never retries a write.

The `capacity create` / `scale` / `renew` / `delete` and `overflow` commands are high-risk writes: preview with `--dry-run`, then confirm with the runtime-injected `--yes`. Each write is submitted once and never auto-retried; HTTP 200 is not success, so pass `--wait` or check `operation_status`. Scale values are one instance's absolute kTPM, not deltas or ModelCode totals, and zero is not release. `capacity delete` releases an instance but keeps the ModelCode; active prepaid instances cannot be DELETEd — run `capacity unsubscribe --instance-id <id>` to get the billing console refund link and finish there (no API exists for refunds). Release is confirmed by `deleted=true`. `overflow` applies to the whole ModelCode, not one instance. There is no estimator or standalone auto-renewal endpoint — renewal settings ride along purchase/scale/renew.

Flags, usage, and examples: see [`reference/`](reference/index.md) or `bl <command> --help` — do not guess flags.

## Quick examples

```bash
bl dataset validate --file train.jsonl
bl dataset upload --file train.jsonl
bl finetune text create --base-model qwen3-8b --training-type sft-lora --datasets file-xxx
bl finetune watch --job-id ft-xxx
bl finetune export --job-id ft-xxx --checkpoint ckpt-3 --model-name my-qwen-sft
# Optional quantization (Beijing region; full-parameter fine-tuned models only)
bl finetune compress templates --model my-qwen-sft
bl finetune compress create --model my-qwen-sft --template-id quant-xxx --suffix int4
bl finetune compress watch --job-id quant-xxx --follow
bl deploy text create --model-name <my-qwen-sft-or-quantized-output> --display-name my-svc
```

## Common hand-offs

软 hand-off（按 skill **名**；已安装则 Read，否则 `--help` / 提示 `bl skill init`）：

- After deployment, try the model or generate content → skill `bailian-gen` (media) or `bl text chat` (fallback: `bl image\|video\|text --help`).
- Unsure which base model to pick → `bailian-model-recommend` / `bl advisor recommend`.
- Training quota / usage questions → skill `bailian-cli` (fallback: `bl quota` / `bl usage --help`).

## references

- [bailian-protocol](../bailian-protocol/SKILL.md) — shared protocol (install via `bl skill init`)
- [reference/](reference/index.md) — command details

Similar skills

Everything for AI and ML →

finetune Seth Hobson Runs the whole fine-tuning lifecycle, from building the eval harness to exporting a checkpoint. Each phase is gated by the artifact the previous one produced. CX Codex ★ 41,729repo stars tao-finetune-cosmos-reason NVIDIA Corporation The skill plans and runs Cosmos3 fine-tuning through Cosmos Framework or Cosmos-RL, validating every input before training starts. CX Codex +1 1,600installs senior-ml-engineer Alireza Rezvani Guides production ML deployment, MLOps pipelines, RAG systems and LLM API integration. Use it when a project needs monitoring, rollout checks or request cost controls. 915installs ai-product-strategy RefoundAI Guides product decisions for AI features, helping you choose between RAG and fine-tuning and plan for non-deterministic model behavior. CX Codex +1 2,700installs eval-harness-first Seth Hobson Builds the evaluation harness before fine-tuning with goldens, graders, judge calibration, and a base-model baseline. Useful for tracking regressions. CX Codex 2,600installs axolotl Orchestra Research Walks your agent through Axolotl training runs: YAML configs, LoRA and QLoRA, DPO, KTO, ORPO and GRPO methods, dataset formats. CX Codex +1 960installs

From the same set7

bailian-cli Skill for managing Aliyun Bailian and DashScope resources via the bl CLI. It routes app calls, quota lookups, and knowledge base queries. CX Codex +1 18,800installs bailian-protocol Shared execution protocol for the Aliyun Bailian bl skill family. It sets confirmation rules, provider routing, and safeguards for high-risk operations. CX Codex +1 9,200installs bailian-gen Generates and edits images, video, and audio through Bailian, with media understanding tools too. Use it when an agent needs to create, transcribe, or inspect media files. CX Codex +1 9,200installs bailian-managed-agent This skill drives Anthropic Bailian agents, environments and deployments from agents.yaml. Useful when infrastructure changes should go through a plan, not by hand. CX Codex +1 9,200installs bailian-web-search This skill picks the search path for CLI bl: the model's built-in search or Bailian MCP. Useful when you work with Alibaba Bailian and keep mixing up keys. CX Codex +1 3,100installs bailian-sandbox A skill that drives Alibaba Bailian sandbox instances and templates from the terminal: create, connect, pause, release. CX Codex +1 812installs