---
title: "Start Quantization"
method: POST
path: "/v1/developer/{developer_id}/quantize"
tags: ["Developer Management", "Quantization", "Quantization"]
---

# Start Quantization

`POST /v1/developer/{developer_id}/quantize`

## Request

- Base URL: `https://api.oneinfer.ai` (the document declares no server; this is the origin it was published from)
- URL: `https://api.oneinfer.ai/v1/developer/{developer_id}/quantize`
- Auth: HTTP bearer

Start a new quantization job on an existing GPU instance.

## Path parameters

- `developer_id` string, required

## Request body

- StartQuantizationRequest: Request body for starting a quantization job.
  - `model_source` 'huggingface' | 'catalog', required
  - `hf_repo` string, nullable
  - `model_id` string, nullable
  - `scheme` string, required
  - `dataset` string
  - `prompt` string
  - `instance_id` string, required
  - `provider` string, required
  - `region_override` RegionOverrideRequest: Mixed-precision request: quantize one tensor region to a different scheme than the rest of the model.
    - `region` 'embeddings' | 'attention' | 'attention_qkv' | 'attention_o' | 'mlp' | 'mlp_gate_up' | 'mlp_down' | 'output', required: Selectable GGUF tensor regions for mixed-precision (per-layer) quantization. Maps to llama.cpp GGUF tensor naming (token_embd.weight, blk.N.attn_*, blk.N.ffn_*, output.weight) via TENSOR_REGEX_BY_REGION in quantization_deploy.py, passed to `llama-quantize --tensor-type <regex>=<TYPE>`.
    - `override_scheme` string, required
    - `layer_start` integer, nullable
    - `layer_end` integer, nullable

## Response `202`

Successful Response

- unknown

## Other responses

- `422`: Validation Error

---

[API](https://skmtc.dev/oneinfer/apis/oneinfer-apis.md) · [All operations](https://skmtc.dev/oneinfer/apis/oneinfer-apis/llms.txt) · [OpenAPI document](https://skmtc.dev/oneinfer/apis/oneinfer-apis/revisions/a49f22311a8f?raw)