Qwen3.5 122B A10B

Q4_K · huihui-ai · Measured configuration

80.72 / 100 Mango weighted score · 120.76 min matched workload · 11/12 mechanical agent passes

The eight-shard Q4_K package of huihui-ai's abliterated Qwen3.5-122B-A10B. All eight selected shards retain their original Q4_K-GGUF/ paths.

Benchmark results · Files · Download · Runtime · Provenance · Full collection

Join Infinity on Discord

Get model support, request models and quants, discuss benchmarks, and follow new releases and model improvements.

Join the Infinity community

All model tests, per-question scores, experiment history and scoring methods

Qwen effort study — separate full-404 cohort: medium 82.87/100 versus xhigh 84.41/100; paired difference +1.54 points (95% interval [-1.71, +4.82]). 404 collected questions, 403 valid questions graded per effort; one invalid reference annotation excluded symmetrically. The historical 202-question cross-model comparison below is unchanged. Study, timing and blind subjective results.

Preserved and documented by isilicon / Mango. The original model, modification and quantization remain the upstream creators' work. This repository adds attributed preservation and exact-configuration evaluation, with the selected source bytes unchanged.

At a glance

GB is decimal and GiB is binary. Payload size is disk/storage size, not a claim that the model fits in the same amount of VRAM. Runtime, KV cache, working memory and companions add their own requirements.

Mango evaluation

It passed all 20 strict IFEval prompts and 18/20 local coding tests. Eleven of the twelve agent cases passed the mechanical heuristic; the failed case remains included.

The 80.72 weighted score is close to Qwen27's 80.64, while 120.76 workload minutes is about five times Qwen27's time. This exact tested configuration has not demonstrated a quality gain that offsets that cost.

Domain scores equally average their component task means. The overall score applies the weights above. EQ is continuous; the 12-case mechanical agent pass rate is only one component of the agentic domain.

Task-level quality and uncertainty

These are small released subsets and local graders. HumanEval/BFCL scores are not the complete official benchmark pipelines. Agent checks cover completion, evidence, permissions and applicable pricing totals; other answer semantics remain unjudged.

Recorded performance

Workload time excludes loading, restarts, wipe downtime and idle gaps. Latency/TTFT summaries cover suite requests only. The reported token rate divides engine usage counts by request time; it is not native decode throughput and token accounting differs across engines. No concurrent-serving result is implied.

Matched cohort comparison

The old mixed GLM row is a historical, internally capped baseline whose exact weights are unavailable. Other rows also differ in quant, engine, context and effort. Point ranks do not settle the quality ordering.

Paired score differences

Intervals use 10,000 paired task-stratified bootstrap resamples, seed 41, with fixed task/domain weights. An interval containing zero does not establish a quality winner. See methodology for scope and definitions.

Download the current full numerical report and methodology. The original compact evaluation is retained as the earlier preservation snapshot; its model score is unchanged, while the current report includes the expanded cohort. The numerical report contains aggregate statistics and hashes; no prompts or generated answers are published.

Runtime and compatibility

Download every shard before loading the first. Mango used llama.cpp with automatic GPU fitting/offload and Q8 KV caches. A benchmark on a 96GB-class GPU with host RAM does not establish that all weights fit in VRAM.

32,768-token configured test context; native engine-default effort in the saved responses.

The local evaluation covers text/tool tasks. It does not establish the full advertised context ceiling, image/audio quality or every optional speculative-decoding path. Publisher capabilities and Mango-verified behavior are separate scopes.

Companion artifacts

No separate companion mirror is declared in this selected manifest. This does not imply that every optional upstream projector, draft head or multimodal path is included; consult the exact file list.

Files and integrity

Full paths, byte sizes and complete SHA-256/Git blob identities are in SOURCE_MANIFEST.json. Available original README/license/notice files are also preserved. MIRROR_STATUS.json records the earlier source-file verification; MODEL_CARD_STATUS.json records this documentation refresh separately.

Download and use

The commands pin the already-verified mirror weight revision. Documentation can advance on main without changing that preserved weight snapshot. Keep original subfolders and all shards.

python -m pip install -U huggingface_hub

hf download isilicon/mango-qwen35-122b-q4 \
  Q4_K-GGUF/Q4_K-GGUF-00001-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00002-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00003-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00004-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00005-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00006-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00007-of-00008.gguf \
  Q4_K-GGUF/Q4_K-GGUF-00008-of-00008.gguf \
  --revision 755644ad14ac7370d0ecc5f1242b4e8b9fa93245 \
  --local-dir ./weights/qwen35-122b-q4

Load example and setup scope

llama-server \
  --model ./weights/qwen35-122b-q4/Q4_K-GGUF/Q4_K-GGUF-00001-of-00008.gguf \
  --host 127.0.0.1 --port 8080 --ctx-size 32768 \
  --jinja --flash-attn on --cache-type-k q8_0 --cache-type-v q8_0

This is a minimal text-serving example for a compatible build, not a replay of every benchmark fitting/offload flag. The recorded configuration and engine caveats define the tested scope.

Provenance and license

huihui-ai supplies the modified checkpoint and GGUF release; the declared base is Qwen/Qwen3.5-122B-A10B. The larger parameter count is an architecture description, not evidence of a better local quality/time tradeoff.

Original notices, base-model conditions and the publisher's stated limitations still apply. Downloadable preservation does not add new license rights. Source-described refusal modification and publisher-side benchmark claims have not been independently re-measured by this Mango task cohort.

Original publisher documentation

The unchanged upstream README and pinned source card retain the original usage guidance, technical discussion, attribution, caveats and citations.

Read the complete original publisher card

Source reference. The following is the original publisher's documentation. Its other quants, full-model files, context claims and benchmark tables can exceed the selection in this mirror. Relative asset/file URLs are qualified to the pinned original source; publisher measurements remain separate from Mango measurements.

huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF

This is an uncensored version of Qwen/Qwen3.5-122B-A10B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Download and merge

Use the llama.cpp split program to merge model (llama-gguf-split needs to be compiled.),

huggingface-cli download huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF --local-dir ./huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF --token xxx


llama-gguf-split --merge huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF/Q4_K-GGUF/Q4_K-GGUF-00001-of-00008.gguf huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF/ggml-model-Q4_K.gguf

chat_template-vl-think.jinja

We have added a new file named chat_template-vl-think.jinja, which comes from the path huihui-ai/Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated. This template file supports the think mode.

The new file chat_template-vl.jinja is more compatible with using Tool Calling in llama-server, especially when opencode and oh-my-opencodeis involved.

This will help prevent 500 error messages from occurring in the llama-server.

llama-server -m huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF/ggml-model-Q4_K.gguf --port 8080 --host 0.0.0.0 -c 262144 --chat-template-file huihui-ai/Huihui-Qwen3.5-122B-A10B-abliterated-GGUF/chat_template-vl-think.jinja

The following are the relevant configurations for openconde.json used in a Docker environment.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "llama-server": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "llama-server",
      "options": {
        "baseURL": "http://host.docker.internal:8080/v1"
      },
      "models": {
         "Huihui-Qwen3.5-122B-A10B-abliterated-Q4_K": {
          "name": "Huihui-Qwen3.5-122B-A10B-abliterated-Q4_K",
          "tools": true,
          "reasoning": true,
          "options": {
            "num_ctx": 262144
          }
        }
      }
    }
  }
}

Usage Warnings

• Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs.

• Not Suitable for All Audiences: Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security.

• Legal and Ethical Responsibilities: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences.

• Research and Experimental Use: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications.

• Monitoring and Review Recommendations: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content.

• No Default Safety Guarantees: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use.

Donation

If you like it, please click 'like' and follow us for more updates.
You can follow x.com/support_huihui to get the latest model information from huihui.ai.

Your donation helps us continue our further development and improvement, a cup of coffee can do it.

• bitcoin（BTC):

  bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge

• Support our work on Ko-fi (https://ko-fi.com/huihuiai)!

Preservation record

Selected source bytes are verified against pinned metadata; the original eight graded mirrors also passed the previously recorded anonymous 1 MiB weight-prefix checks. This is not a full independent-device restore/load test. Hugging Face repository copies share a provider and do not guarantee permanent storage.

Collection · Source manifest · Original documentation