---

license: apache-2.0

base_model: LessThanThreeAI/Qwen3.8-27B-Humanlike-Chat

base_model_relation: quantized

library_name: gguf

pipeline_tag: text-generation

tags:

- gguf

- llama.cpp

- qwen3.8

- conversational

- roleplay

- creative-writing

- character

- humanlike

- uncensored

- sillytavern

---

Qwen3.8-27B Humanlike Chat GGUF

> A Qwen3.8-27B checkpoint that is unmistakably human in conversation.

Uncensored | Humanlike dialogue | Roleplay | Personal chat | Interactive fiction | Local GGUF

Example of interaction with Humanlike in role of tavern keeper

user
I am local - Thorsten's the name

Brannoc
Thorsten. Yeah, you're local, I can tell.
You walk the road like you know every mud patch on it.
Well, welcome home, then.
The plum wine's good. Made it last autumn.
Tastes like a bit of autumn still, you know?
What brings you by tonight, besides the weather, if I may ask?

user
old enemy...

Brannoc
Old enemy. Hm.
You need a room that can be locked from the inside,
or you just want to sleep?

> WIP. Improvements to reasoning mode support are pending

This is not a standard fine-tune built around general benchmark performance. It is a behavior adaptation trained on what a person actually said next across long-running conversations.

That changes more than vocabulary and sentence structure. It changes what the model notices, what it leaves unanswered, when it jokes, when it pushes back, how it follows subtext, and how much it chooses to say.

Built for roleplay, characters, personal chat, companion experiences, interactive fiction, and any application where the model should feel like a participant rather than an assistant.

Uncensored by design

This model inherits the reduced-refusal behavior of the Huihui Qwen3.8-27B Abliterated checkpoint.

That parent was chosen deliberately. Refusal-related directions can affect more than which topics a model will discuss. They can also influence tone, phrasing, response structure, conversational choices, and how readily a character follows the natural direction of an interaction.

Starting from the abliterated parent was intended to give the adaptation more room to learn natural human behavior without pulling every conversation back toward assistant-style compliance.

What humanlike means here

• Responds to the intent that matters instead of mechanically addressing every detail
• Follows mood, subtext, relationship dynamics, callbacks, and topic changes
• Takes initiative without hijacking the conversation
• Jokes, disagrees, pushes back, and leaves things unsaid
• Avoids reflexive questions, lists, summaries, and canned reassurance
• Uses short reactions when a short reaction is what a person would actually send
• Keeps characters conversational instead of turning them into assistants wearing costumes

It does not need a persona prompt to activate this conversational style. Use a character card or system prompt when you want a specific identity, setting, or relationship.

Start here

Choose the highest quant that fits while leaving memory for context and the KV cache.

| Hardware | Download | Size | Recommendation |

|---|---|---:|---|

| 32 GB+ VRAM | Q8_0 | 28.60 GB | Highest practical quality |

| 24 GB+ VRAM | Q6_K | 22.08 GB | Best quality and memory balance |

| 20 GB or CPU offload | Q4_K_M | 16.55 GB | Lower-memory local use |

| 64 GB+ VRAM | BF16, two shards | 53.81 GB | Maximum fidelity and requantization |

VRAM needs rise with context size and KV-cache settings. Partial CPU offload can run each quant with less GPU memory.

SillyTavern quickstart

Start a local llama.cpp server:

llama serve -hf LessThanThreeAI/Qwen3.8-27B-Humanlike-Chat-GGUF:Q8_0 \
  --ctx-size 32768 \
  --parallel 1 \
  --n-gpu-layers all \
  --jinja \
  --chat-template-kwargs '{"enable_thinking":false}'

In SillyTavern, connect with:

API: Text Completion
Backend: llama.cpp
URL: http://127.0.0.1:8080
Context: 32768
Response length: 512
Temperature: 0.7
Top P: 0.8
Top K: 20
Presence penalty: 1.5
Repetition penalty: 1.0
DRY: disabled initially
Smoothing: disabled initially
Manual stop strings: unset

For the most reliable personal-chat experience, use non-thinking mode (enable_thinking=false). Thinking mode is experimental and may produce incomplete replies on some runtimes.

Qwen3.8 has a native context length of 262,144 tokens. Start at 32,768 for roleplay, then increase to 65,536, 131,072, or 262,144 as available memory permits. Larger contexts require more memory for the KV cache.

<!--

REVIEW TODO BEFORE PUBLICATION

Add three public-safe examples generated from the exact final Q8_0 artifact:

1. A personal conversation with mixed intent.
2. A character-roleplay exchange involving subtext or disagreement.
3. Informal banter showing brevity, initiative, and no assistant boilerplate.

For each example, record:

• Public model revision
• Quant
• Backend and build
• Context size
• Character/system prompt
• Sampler
• Thinking mode
• Seed

-->

Natural-chat results

In retained natural-chat replays:

• 0 detected slop trigrams across 590 turns
• 0 detected "not X but Y" constructions across 590 turns
• 0 list-form replies in 590 turns
• Median reply length fell from 427 to 32 characters versus the base replay
• Reflexive questions fell from 27/29 turns to 3/29
• The tuned model gave the shorter reply on 300 paired turns

The result is a model that can hold a realistic conversation instead of constantly trying to manage one.

Why it feels human

The model was trained on 139,845 real messages across 1,396 sessions from long-running one-to-one conversations. Those conversations produced 7,006 training examples and 194 held-out validation examples.

Each example presents the conversation so far and trains only on what one speaker actually said next. Prompt and history tokens do not contribute to loss.

That teaches more than writing style. It teaches conversational timing, continuity, implied intent, relationship dynamics, callbacks, mood, and the human choice to leave some things unanswered.

Great for

• Character roleplay and multi-character scenes
• Uncensored SFW and NSFW roleplay
• Personal or companion-style chat
• Interactive fiction and dialogue-heavy games
• Creative improvisation
• NPCs with a distinct conversational personality
• Social and relationship-driven simulations

The adaptation

This is a rank-256 adaptation across all 496 language modules of Qwen3.8-27B: 1.87 billion trainable parameters, 992 LoRA tensors, 7,006 training examples, and 158,085 supervised response tokens.

It was trained from the abliterated Huihui Qwen3.8-27B parent, merged into text-only BF16, and then quantized directly from the merged BF16 GGUF.

FP32 LoRA adapter

The unmerged step-863 FP32 adapter is also available: 7.47 GB, rank 256, alpha 32, 992 F32 tensors. This is a GGUF LoRA for llama.cpp's --lora option, not a PEFT safetensors adapter or a standalone model.

Use it with an unadapted GGUF conversion of huihui-ai/Huihui-Qwen3.8-27B-abliterated at revision d42ca8978c5a66e92c3446d46e8adfe03ef692ff. Do not apply it to the merged models in this repository: they already contain the adapter. Keep thinking disabled for the verified path; reasoning remains experimental. The exact adapter hash, matched base and validation scope are in artifact-manifest.json.

Base compatibility

This adapter was trained on Huihui Qwen3.8-27B Abliterated, which is the best-supported match.

You can also experiment with stock Qwen3.8-27B and other abliterated derivatives of the same 27B architecture, including compatible GGUF quantizations supported by your runtime. FP32 describes the adapter's precision; it does not require an FP32 base model.

Other bases and quantization levels may change conversational style, refusal behavior, and reliability. These combinations are experimental, not individually verified. Matching architecture, tensor layout, and tokenizer are required; this is not an adapter for arbitrary Qwen models or other model sizes.

Applying the adapter to a quantized base is not numerically identical to quantizing the already-merged model. The published merged-quant checks do not establish quality for every external base-plus-adapter combination. Apply it only to an unadapted base, never to models that already contain this adapter.

Technical specifications

| Item | Specification |

|---|---|

| Base model | huihui-ai/Huihui-Qwen3.8-27B-abliterated, based on Qwen/Qwen3.8-27B |

| Architecture | Dense 27B text model, 64 language layers |

| Adaptation | Rank 256, alpha 32, 496 language modules, 992 LoRA tensors |

| Trainable parameters | 1,867,644,928 |

| Context length | 262,144 tokens native; adaptation trained at 4,096 tokens |

| Precision and formats | Merged BF16, GGUF BF16, Q8_0, Q6_K, and Q4_K_M |

| Model size | 16.55 GB to 53.81 GB |

| Inference frameworks | llama.cpp, SillyTavern, Ollama, LM Studio, Jan, KoboldCpp |

| Measured speed | Q6_K: 52.49 tokens/s on RTX PRO 6000 Blackwell Max-Q |

| Modality | Text only; vision and MTP/NextN tensors excluded |

| License | Apache-2.0 |

Quantization and integrity

All quants were created directly from the merged BF16 GGUF using llama.cpp@8144f3192e5a3131cd043f284525e6ceebf82d0f. No quant was produced from another quant.

Exact file hashes are in SHA256SUMS.

Lineage

Qwen/Qwen3.8-27B
  -> huihui-ai/Huihui-Qwen3.8-27B-abliterated
  -> rank-256 step-876 Humanlike Chat LoRA
  -> merged text-only BF16
  -> BF16 / Q8_0 / Q6_K / Q4_K_M GGUF

Hugging Face and Transformers may display the architecture as qwen35 or qwen3_5_text. That is Qwen3.8's internal implementation identifier. This release is Qwen3.8-27B.

Share your results

Tried it in SillyTavern, KoboldCpp, LM Studio, or another local frontend?

Post your character, quant, context size, sampler, and best conversation examples in the Community tab. Strong presets and reproducible examples may be added to this card.

License

Apache-2.0, inherited from the upstream Qwen and Huihui releases.