---

license: apache-2.0

language:

- ja

- zh

tags:

- asmr

- translation

- japanese

- chinese

- gguf

- long-context

base_model: unsloth/Qwen3.5-9B

quantized_by: mmis1000

pipeline_tag: text-generation

---

asmr-qwen3.5-9b-zh-cn-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.

Standard Mode

The traditional output format where only the translated text is returned.

Available Quantizations

| Quantization | Filename | Size | Description |

|---|---|---|---|

| q4_k_m | asmr-qwen3.5-9b-zh-cn-gguf-v0.2-q4_k_m.gguf | 5.2 GB | Good balance of quality and size |

| q6_k | asmr-qwen3.5-9b-zh-cn-gguf-v0.2-q6_k.gguf | 6.9 GB | Higher quality, moderate size |

| q8_0 | asmr-qwen3.5-9b-zh-cn-gguf-v0.2-q8_0.gguf | 8.9 GB | Near-lossless quality |

| bf16 | asmr-qwen3.5-9b-zh-cn-gguf-v0.2-bf16.gguf | 16.7 GB | Full BF16, no quantization loss |

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

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>, ...], "text": "<简体中文>", "start": <最早ms>, "end": <最晚ms>}

字幕版权行：{"ids": [<n>], "text": null, "start": <ms>, "end": <ms>}
每个输入id必须恰好出现在一个输出项中。

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], "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-gguf-v0.2-q4_k_m.gguf -c 8192 --port 8080

llama-cli

llama-cli -m asmr-qwen3.5-9b-zh-cn-gguf-v0.2-q4_k_m.gguf -c 8192 -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
      },
      "text": {
        "anyOf": [
          {
            "type": "string"
          },
          {
            "type": "null"
          }
        ]
      },
      "start": {
        "type": "integer"
      },
      "end": {
        "type": "integer"
      }
    },
    "required": [
      "ids",
      "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, q_proj, k_proj, down_proj, v_proj, up_proj
• Locale: zh-cn (Simplified Chinese)
• Mode: Standard Mode
• Max sequence length: 8192
• Precision: bf16

Version 0.2 — Native 8k Continuation

This release continues the v0.1 Simplified Chinese standard 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: 78d239dcde5030af979030ef5ea5e7d515638b1f46117fae9934f9a184c7ec59.

Pinned base revision: 005429cee5cb648998cf2b70eebdd83175989c9a.

Validation Scope

This variant received a small held-out semantic check rather than the full 200-row matrix: 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 found five of six rows semantically usable. The strongest long rows preserved coherent scene progression over 60+ source segments. The single failed row contained catastrophically corrupted source ASR, including eight repeated おやすみなさい segments; both model output and reference contained content unsupported by the literal source.

Context Budgeting

8192 is the total runtime context, not a safe input-only budget. Reserve room for generated JSON. Corrupted or repetitive ASR should be split into shorter windows rather than concatenated to the limit.

Known Limitations

• This variant did not receive the full 100-row 4k plus 100-row 8k validation used for zh-tw-echo.
• Catastrophically corrupted source ASR can cause invented content even when output structure remains valid.
• Local mistranslations remain on ambiguous or malformed ASR fragments.
• 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.