asmr-qwen3.5-9b-zh-cn-echo-gguf-v0.2

GGUF quantizations of a fine-tuned model for translating Japanese ASMR transcriptions (ASR/Whisper output) into Simplified Chinese.

The model normalizes imperfect audio transcriptions, applies domain-specific glossaries, and translates character dialogue while retaining emotion and nuances.

This variant preserves the upstream Qwen3.5 MTP / speculative-decoding head in GGUF format so it can be used with MTP-capable llama.cpp builds.

Echo Mode

The model echoes the source Japanese text in an "input" field and records applied terms in a per-entry "glossary" object alongside the target translation. This provides an explicit source anchor that can reduce omitted or drifted segments, but it does not guarantee immunity to long-context repetition or noisy-ASR failures.

Available Quantizations

Prompt Example

将以下日语ASMR逐字稿翻译成简体中文。

音轨：track01_示例音轨
场景说明：主角与青梅竹马在校园下午的对话...

术语表（请严格使用zh栏位的译名）：
{
  "cvs": [],
  "characters": [],
  "terms": [{"ja": "放課後", "zh": "放学后"}]
}

翻译前请静默修正以下Whisper识别错误：
- 重复片语（连续3次以上且无变化）：仅保留一次
- 错字／同音异字：依上下文修正
- 字幕版权行（字幕：／翻訳：／QQ／LINE水印）：text设为null
- 错误专有名词：依术语表修正

翻译规则：
- 呻吟与气息声（あ、ん、はあ）→ 自然对应（啊、嗯、哈、呼）
- 拟声词：日语形式翻译（パンパン→啪啪）；中文形式保留原样
- 保留角色语气与口吻
- text字段只输出译文，不加注释或括号说明
- input字段为ids所对应的原始日文片段
- glossary字段只记录该输出项实际使用的日文→中文术语；没有时输出空对象

输入：逐字稿JSON数组 — {"id": <n>, "text": "<日文>", "start": <ms>, "end": <ms>}

输出：将连续构成同一句话的片段合并，JSON数组格式：
{"ids": [<n>, ...], "input": "<合并后的原始日文，以空格连接>", "glossary": {"<日文>": "<简体中文>"}, "text": "<简体中文>", "start": <最早ms>, "end": <最晚ms>}

字幕版权行：{"ids": [<n>], "input": "<原始日文>", "glossary": {}, "text": null, "start": <ms>, "end": <ms>}
每个输入id必须恰好出现在一个输出项中。
input字段为ids所对应的原始日文片段以空格连接，glossary记录该项实际使用的术语，text为其简体中文翻译。

逐字稿：
[
  {"id": 1, "text": "ねぇ、放課後、", "start": 3000, "end": 5000},
  {"id": 2, "text": "一緒に帰らない?", "start": 5000, "end": 7000}
]

Example Output:

[{"ids": [1, 2], "input": "ねぇ、放課後、 一緒に帰らない?", "glossary": {"放課後": "放学后"}, "text": "呐，放学后，要不要一起回去？", "start": 3000, "end": 7000}]

Usage

llama-server

llama-server -m asmr-qwen3.5-9b-zh-cn-echo-gguf-v0.2-q4_k_m.gguf -c 8192 --port 8080

llama-cli

llama-cli -m asmr-qwen3.5-9b-zh-cn-echo-gguf-v0.2-q4_k_m.gguf -c 8192 -p "<your prompt>" -n 2048

llama-cli with MTP

llama-cli -m asmr-qwen3.5-9b-zh-cn-echo-gguf-v0.2-q4_k_m.gguf -c 8192 --spec-type draft-mtp --spec-draft-n-max 6 -p "<your prompt>" -n 2048

Structured Decoding (Recommended)

This model outputs JSON arrays. Using structured decoding (e.g. GBNF grammar or JSON schema constraints) avoids wasted computation on malformed output and guarantees valid JSON on every generation.

JSON Schema:

{
  "type": "array",
  "items": {
    "type": "object",
    "properties": {
      "ids": {
        "type": "array",
        "items": {
          "type": "integer"
        },
        "minItems": 1
      },
      "input": {
        "type": "string"
      },
      "glossary": {
        "type": "object",
        "additionalProperties": {
          "type": "string"
        }
      },
      "text": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ]
      },
      "start": {
        "type": "integer"
      },
      "end": {
        "type": "integer"
      }
    },
    "required": [
      "ids",
      "input",
      "glossary",
      "text",
      "start",
      "end"
    ],
    "additionalProperties": false
  },
  "minItems": 1
}

Supported by llama.cpp (--json-schema), vLLM, and outlines.

Training Details

• Base model: unsloth/Qwen3.5-9B
• Method: LoRA (r=16, alpha=16)
• Target modules: o_proj, gate_proj, k_proj, down_proj, v_proj, up_proj, q_proj
• Locale: zh-cn (Simplified Chinese)
• Mode: Echo Mode
• Max sequence length: 8192
• Precision: bf16

Version 0.2 — Native 8k Continuation

This release continues the v0.1 Simplified Chinese echo adapter directly to a maximum training sequence length of 8,192 tokens. It does not stack adapters and does not use RoPE scaling or YaRN.

The deterministic continuation consumed 800 training draws: 520 coherent same-track, 120 far-context/glossary-recall, and 160 ordinary replay draws. The published adapter reached absolute step 200. Adapter weights SHA-256: 8000e9e54726e1943fd0c63c7f1249a351f462c2902274539e9245f483443740.

Pinned base revision: 005429cee5cb648998cf2b70eebdd83175989c9a.

Validation Scope

This variant received a small held-out semantic check: three ordinary rows and three long rows, each from a distinct locked-test family and with zero source overlap with the continuation ledger. All six grammar-constrained Q8 requests completed with exact ID coverage.

Manual reading classified two rows as pass, two as marginal, and two long rows as major failures. One long noisy-ASR row developed repeated speaker-prefix/template contamination. Another confused an earlier weapon scene with a later ear-cleaning scene.

The three exact original 4k constituent prompts from the repetition failure were then evaluated separately with the identical Q8, grammar, wrapper, seed, and decoding settings. The severe speaker-prefix collapse did not reproduce at 4k. This points to cross-window contamination when noisy/repetitive material is concatenated into a long prompt.

Context Budgeting

8192 is the total runtime context, not a safe input-only budget. Reserve room for generated JSON. Do not fill the entire context with the prompt and expect a non-trivial response.

Recommended Use

Treat long-context use of this variant as experimental. Prefer shorter windows around 4k for noisy ASR, and inspect or retry outputs when concatenating multiple scenes. Structured decoding guarantees the schema but cannot prevent semantically repeated or contaminated text.

Known Limitations

• This variant did not receive the full 100-row 4k plus 100-row 8k validation used for zh-tw-echo.
• Long concatenated noisy ASR can trigger speaker-prefix repetition and cross-scene contamination.
• Shortening to original 4k windows removed the severe collapse, but did not make badly transcribed lines semantically reliable.
• There is no complete controlled v0.1-versus-v0.2 final comparison.
• Validation used Q8_0; other quantizations were not independently quality-scored.

Content Notice

The training domain includes adult ASMR dialogue and may produce sexually explicit text. This model is intended for transcription translation and subtitle-processing workflows.

MTP-preserving export

This revision preserves and verifies all 15 upstream MTP tensors against the pinned base checkpoint after the LoRA merge. The native-8192 adapter is unchanged; no retraining was performed. The MTP head itself is not fine-tuned. Earlier revisions of this v0.2 repository omitted MTP. Filenames and translation weights remain compatible; use a fresh repository revision to avoid cached non-MTP files. All four quantizations have verified MTP metadata and tensor inventory. Q4_K_M was GPU-smoke-tested with llama.cpp b9247, context 8192, and --spec-type draft-mtp --spec-draft-n-max 3. This is a runtime smoke test, not a new quality evaluation or a speedup guarantee. See mtp-preservation.json and mtp-release-verification.json for provenance and checks.