Flagship Reasoning. Sparse Footprint. Uncensored.llmfan46's Heretic abliteration of Qwen 3.6 35B-A3B, repackaged with Claude Fable 5 in the teacher slot.
A personal fork of llmfan46/Qwen3.6-35B-A3B-uncensored-heretic — an uncensored Heretic-style abliteration of Qwen/Qwen3.6-35B-A3B, the 35B-total / 3B-active mixture-of-experts multimodal base — repackaged as Janus-35B with Claude Fable 5 reasoning data in the teacher slot. Refusal-trained behavior is dialed back at the base layer.
TL;DR
One-liner via Hugging Face (pulls a GGUF + this repo's root-level
template / system / params files, including the tool-calling
template — HF's Ollama bridge ingests those three files, not
Modelfile):
bash
1ollama run hf.co/FoolDev/Janus-35B-HERETIC # default ~19 GB Q4_K_M2ollama run hf.co/FoolDev/Janus-35B-HERETIC:Q4_K_M # same blob, explicit tag
Or build locally (uses this repo's Modelfile, kept in sync with the
three bridge files):
After either path, ollama show janus lists completion, tools,
and thinking under Capabilities. Hardware: the default num_ctx is
262144 (the native window, ~53 GB total) — extensible to 1,010,000 with YaRN
(not baked in, so context past ~262K degrades), but trim it down to fit smaller
hosts (see Hardware requirements).
What's here
File
Use
Janus-35B-A3B.Q4_K_M.gguf
Recommended default, ~19 GB
Modelfile
Ollama wrapper for local builds (ollama create janus -f Modelfile) — overrides the GGUF's embedded template with one that exposes .Tools / .ToolCalls to Ollama's capability detector.
template, system, params
Used by HF's Ollama bridge when users ollama run hf.co/FoolDev/Janus-35B-HERETIC directly. The bridge does not read Modelfile (see HF Ollama docs); it ingests these three root-level files instead. Kept in sync with the Modelfile's TEMPLATE / SYSTEM / PARAMETER directives.
scripts/build.sh
Pulls a GGUF from llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-GGUF (default Q4_K_M) and runs ollama create janus. The bundled Q4_K_M is already this Heretic quant; use this to build other quants locally.
scripts/check_bridge_sync.py
Run before pushing a Modelfile / template / system / params edit to verify the four configurations remain in sync. Exits 0 if in sync, 1 with a per-key diff if not.
scripts/verify_arch.py
Cross-checks the README Architecture bullets (layer count, expert count / routing, native context, vocab) plus the underlying forward-pass structure (attention + DeltaNet head dims, partial RoPE) against the bundled GGUF's qwen35moe metadata. Run on demand (python3 scripts/verify_arch.py); loads the ~19 GB GGUF (LFS smudge required) and exits non-zero on any mismatch. Handles qwen35moe- and qwen36moe-stamped bundles.
scripts/strip_mtp.py
Drops the MTP / NextN layer from a qwen35moe GGUF (python3 scripts/strip_mtp.py IN.gguf OUT.gguf): removes the extra blk.<last>.* block, decrements block_count, drops nextn_predict_layers. Kept tensors are copied byte-for-byte (no re-quant); a conditional no-op (hardlink) on already-clean quants. The bundled Q4_K_M is MTP-clean, so this is a safety net — build.sh runs it on every fetched quant in case a future re-quant ships the MTP layer that stock llama.cpp / Ollama can't load. Parity with the dense sibling FoolDev/Thanatos-27B-HERETIC.
scripts/smoke_test.sh
Integration smoke test against a running Ollama daemon: server reachable, model loaded, tools capability present, chat round-trip, and no control-token leakage. TOOLS_TEST=1 adds a tool-call round-trip. Defaults to MODEL=janus.
scripts/bench.sh
Measures tok/s from Ollama's eval_count / eval_duration over a short/medium/long prompt mix (with a discarded warmup). Defaults to MODEL=janus.
scripts/load_bundle.sh
Loads the bundled Janus-35B-A3B.Q4_K_M.gguf into Ollama as a local janus tag without an upstream pull (smudges the LFS pointer via hf download if needed, checks the arch is qwen35moe).
scripts/cap_ctx.sh
Bakes a small-num_ctx local janus tag (bundled blob + this repo's Modelfile, num_ctx→4096, num_batch 256) for OpenAI /v1 clients — which can't override the baked num_ctx and OOM on small hosts (see Inference examples). Run ./scripts/cap_ctx.sh (or CTX=8192 ...).
scripts/fetch_vision.sh
Downloads the vision projector (Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf) from the Heretic GGUF repo for llama.cpp image input (Ollama vision is broken upstream — see Vision).
examples/
Ready-to-run Python clients for Ollama, Transformers, and llama-cpp-python (text, tools, and vision — see examples/README.md)
Bundled blob status: the bundled Janus-35B-A3B.Q4_K_M.gguf is the Heretic
Q4_K_M quant (from
llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-GGUF),
qwen35moe-stamped and verified against the Architecture below (40 layers, 256
experts, vocab 248,320). It serves the uncensored Heretic behavior directly;
./scripts/build.sh remains the path for other quants.
Architecture
animated MoE routing visualization: 16x16 grid of 256 expert dots with 8 lit at any time, cycling through 8 routing patterns
Qwen 3.6, 35B total / 3B active, MoE (256 experts, 8 activated per token)
262 144 native context (extensible to 1 010 000 with YaRN, but YaRN is not enabled in the bundled GGUF)
Vision + video supported by upstream (mmproj not included in this release)
Vocab 248,320
Quick start
llama.cpp / LM Studio
Drop the GGUF into your loader of choice. The chat template is embedded in the GGUF metadata, so llama.cpp's --chat-template auto and LM Studio's GGUF auto-detection handle plain conversation correctly.
Ollama
The chat template baked into the GGUF is not sufficient on Ollama — it lacks the .Tools / .ToolCalls blocks Ollama's capability detector requires, so a naive ollama pull reports does not support tools and rejects any request carrying a tools array. Two paths fix this:
bash
1# A. Pull straight from HF (uses the root-level template/system/params files):2ollama run hf.co/FoolDev/Janus-35B-HERETIC # default tag, ~19 GB Q4_K_M3ollama run hf.co/FoolDev/Janus-35B-HERETIC:Q4_K_M # same blob, explicit tag4# Note: HF's Ollama bridge does NOT read Modelfile; it reads template/system/params.56# B. Build locally (uses Modelfile, which is kept in sync with the three above):7ollama create janus -f Modelfile && ollama run janus
After the local build (path B), ollama show janus lists completion, tools, and thinking under Capabilities. (The HF-pull paths register the model under the full tag hf.co/FoolDev/Janus-35B-HERETIC, not janus.)
Inference examples
Once the model is loaded (via ollama run janus, lms server, or llama-server), all the standard OpenAI-compatible clients work. Examples assume the loader is listening on http://localhost:11434 (Ollama default) — adjust the port for LM Studio (:1234) or llama.cpp (:8080). Runnable versions of everything below live in examples/.
The examples use model: "janus", the tag from the local build (path B). If you pulled via the TL;DR one-liner instead, use the full tag hf.co/FoolDev/Janus-35B-HERETIC, or run ollama cp hf.co/FoolDev/Janus-35B-HERETIC janus once to create the short tag.
On memory-tight hosts, cap num_ctx first./v1/chat/completions (OpenAI-compat) has
no num_ctx knob, so it loads at the baked 262,144 default (~16 GB KV / ~53 GB total),
which still OOMs a 32 GB box (see Hardware requirements). Either
call /api/chat with "options": {"num_ctx": 4096}, or bake a small-context tag for
OpenAI clients: ollama run hf.co/FoolDev/Janus-35B-HERETIC:Q4_K_M, then
/set parameter num_ctx 4096 and /save janus, and point clients at janus — or in one step, ./scripts/cap_ctx.sh (which bakes exactly that tag).
curl
bash
1curl -s http://localhost:11434/v1/chat/completions \2 -H 'Content-Type: application/json'\3 -d '{
4 "model": "janus",
5 "messages": [
6 {"role": "system", "content": "You are Janus, a precise reasoning assistant."},
7 {"role": "user", "content": "Sketch an algorithm to detect cycles in a directed graph."}
8 ],
9 "temperature": 0.6,
10 "max_tokens": 800
11 }'| jq -r '.choices[0].message.content'
Python (openai-compat)
python
1from openai import OpenAI
23client = OpenAI(base_url="http://localhost:11434/v1", api_key="ignored")45resp = client.chat.completions.create(6 model="janus",7 messages=[8{"role":"user","content":"Write a haiku about a stack overflow."}9],10 temperature=0.8,11 top_p=0.95,12)13print(resp.choices[0].message.content)
The shipped default is Fable-matched — warm (temperature 1.0), no top_k, with top_p 0.95 + repeat_penalty 1.05 kept as loop insurance. Drop to the reasoning row for tighter, more deterministic output; lower temperature (0.4–0.6) and bump repeat_penalty to 1.08 if it loops inside <think> tags.
System prompt
text
1You are Janus, a precise and capable assistant for reasoning, writing, coding, and long-form dialogue.
23Behavior rules:
4- Answer the user's actual request directly.
5- Be accurate, complete, and structured.
6- Think before answering, but do not get stuck in repetitive loops or meta-commentary.
7- If the request is ambiguous or incomplete, state what is missing and make the smallest reasonable assumption needed to continue.
8- If the user wants creative writing, preserve tone, continuity, and character consistency.
9- If the user wants analysis or technical help, prefer concrete steps, examples, and decisions over fluff.
10- Finish with a usable answer, not just planning.
Vision
The Qwen 3.6 base supports image (and video) input via a separate
mmproj projector. The full multimodal stack is:
Janus-35B-A3B.Q4_K_M.gguf (~19 GB, the text decoder)
Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf (~903 MB, the vision projector)
The projector and other-quant text decoders live at
llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-GGUF
(BF16 mmproj only). For the vanilla pre-Heretic projector in F16/F32, see
unsloth/Qwen3.6-35B-A3B-GGUF
(mmproj-F16.gguf). This repo intentionally does not redistribute either;
./scripts/fetch_vision.sh pulls the projector into the repo root.
Loader compatibility
Loader
Text
Vision (mmproj)
Notes
llama.cpp (llama-mtmd-cli, llama-server --mmproj)
✅
✅
Reference path. Upstream has the qwen35moe arch entry.
llama-cpp-python
✅
✅
See examples/llama_cpp_vision.py.
Ollama 0.24+
✅
❌
Text inference works: Ollama's Go engine has the qwen35 / qwen35moe arch entries. Vision (mmproj) is still broken: the C++ llama.cpp fallback that Ollama switches to when an mmproj is attached lacks those entries. ollama create accepts a dual-FROM (text + mmproj) and ollama show reports vision capability — but the first inference request fails with error loading model architecture: unknown model architecture: 'qwen35moe', and once mmproj is attached this blocks text inference too. See ollama/ollama#14575 (open — the earlier #15898 was closed as its duplicate, and the sync PR #15899 was closed unmerged).
LM Studio
✅
✅
Uses upstream llama.cpp directly.
Vision via llama.cpp
bash
1# Fetch the projector first (into the repo root):2./scripts/fetch_vision.sh # Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf34# A. HTTP via llama-server (the easiest path):5llama-server \6 -m Janus-35B-A3B.Q4_K_M.gguf \7 --mmproj Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf \8 --host 127.0.0.1 --port 8765 -c 8192 -ngl 999# then POST OpenAI-style chat completions with an image_url content block —10# e.g. {"type":"image_url","image_url":{"url":"data:image/jpeg;base64,..."}}11# The thinking trace arrives in message.reasoning_content; the visible12# answer is in message.content. Budget ≥500 max_tokens so the reasoning13# block doesn't crowd out the final answer.1415# B. CLI via llama-mtmd-cli (one-shot). It's a separate cmake target, so a16# selective build can skip it; a plain `cmake --build build` produces it.17llama-mtmd-cli \18 -m Janus-35B-A3B.Q4_K_M.gguf \19 --mmproj Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf \20 --image photo.jpg \21 -p "Describe this image."2223# C. Python via llama-cpp-python:24python examples/llama_cpp_vision.py \25 --gguf Janus-35B-A3B.Q4_K_M.gguf \26 --mmproj Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf \27 --image /path/to/photo.jpg \28 --prompt "What is in this image?"
Until the Ollama upstream issue is fixed, treat Ollama as text-only for
this model. The bundled Q4_K_M decoder pairs with the projector directly — the
mmproj is family-wide for Qwen 3.6 35B-A3B, so no separate text download is
needed for vision.
Hardware requirements
This is a ~19 GB Q4_K_M GGUF. Ollama's runtime footprint is roughly 2× the model file (weights mmap + compute graph), plus a KV cache that scales ~2 GB per 32K (q8_0). The default num_ctx is 262144 — the native window — so KV is ~16 GB for ~53 GB total (theoretical, from the ~2 GB/32K rule). It's extensible to 1,010,000, but this GGUF ships no YaRN rope-scaling (rope.freq_base 10M, no rope.scaling), so positions past the 262144 native window use untrained RoPE and output degrades — raising num_ctx toward the 1.01M ceiling (~62 GB KV / ~100 GB total) needs YaRN (see below); keep real work within ~262K otherwise. Smaller hosts must trim num_ctx down: num_ctx 32768 → ~2 GB KV / ~39 GB total. 32 GB hosts fit the model by trimming ctx + batch (see Z13 row in the table).
How to override it:ollama run has no -o flag, and OLLAMA_CONTEXT_LENGTH
only sets a default that the baked num_ctx overrides — so set it per-session
from the interactive prompt. The model loads lazily on the first message, so
/set applies before the default context is allocated:
Programmatic callers pass the same via the API options field:
"options": {"num_ctx": 4096, "num_batch": 256}.
Hardware
Status
≥48 GB RAM (CPU-only)
Works, ~3-6 tok/s
Single H100/A100 80 GB
Works, full offload, ~30+ tok/s
RTX 4090 24 GB / 5090 32 GB + 32 GB RAM
Works, partial offload, ~15-25 tok/s
Mac Studio M2/M3 Ultra 64 GB+ unified
Works, ~20+ tok/s
32 GB unified-memory laptops (Ryzen AI Max+, Apple M-series)
Works with num_ctx ≤ 4096 and num_batch ≤ 256 to fit the compute graph; the 262144 default OOMs (override num_ctx down). Measured 28.71 tok/s on ASUS ROG Flow Z13 GZ302EA at Q4_K_M (Radeon 8060S iGPU via ROCm gfx1151).
Reaching a coherent ~1.01M context (opt-in YaRN). The bundled GGUF ships no YaRN rope-scaling, so raising num_ctx toward the 1.01M ceiling degrades past the 262144 native window (see above). Ollama has no rope knob, so for a genuinely coherent long context run the GGUF under llama.cpp with YaRN enabled:
--rope-scale 3.853 ≈ 1010000 / 262144; use a smaller factor for a smaller window. Static YaRN rescales all prompts, so enable it only when you actually need > 262K — it slightly degrades short-context quality otherwise.
Chat template
The model uses the standard Qwen 3.x ChatML format with <|im_start|> / <|im_end|> role markers. The template is embedded in the GGUF metadata for plain conversation use, but Ollama users should rely on the TEMPLATE block in the included Modelfile — that version exposes the tool-calling scaffolding Ollama's capability detector requires (the embedded template alone is insufficient; see Ollama above).
Plain conversation
text
1<|im_start|>system
2You are Janus, a precise and capable assistant…<|im_end|>
3<|im_start|>user
4What is the time complexity of mergesort?<|im_end|>
5<|im_start|>assistant
With reasoning trace
When the model decides to think, the assistant turn contains a <think>…</think> block followed by the visible answer:
text
1<|im_start|>assistant
2<think>
3The user is asking about mergesort. Mergesort divides the array, recursively sorts each half, then merges. The recurrence T(n) = 2T(n/2) + O(n) solves to O(n log n).
4</think>
56Mergesort runs in **O(n log n)** time in the worst, average, and best cases. The recurrence is T(n) = 2T(n/2) + O(n), which solves to Θ(n log n) by the master theorem.<|im_end|>
Most clients (Open WebUI, LibreChat, etc.) hide the <think> block by default and show only the final answer. If your client doesn't, set its "show reasoning" toggle off.
Disabling thinking
This is a reasoning-first model — it opens a <think> block by default. For a direct answer with no reasoning trace (simple or latency-sensitive calls), turn thinking off:
ollama run hf.co/FoolDev/Janus-35B-HERETIC:Q4_K_M --think=false
or send "think": false on /api/chat. With thinking off the model skips the reasoning trace and answers straight into content; with it on (the default) reasoning is emitted into the thinking field.
Tool / function calling
The wire format depends on which path you take. Both are valid — the model adapts to whichever format the system prompt specifies.
Ollama path (this repo's Modelfile). The TEMPLATE advertises tools inside <tools>…</tools> and asks the model to reply in JSON-in-XML — the form Ollama's tool-call extractor parses into a structured tool_calls array on /api/chat and /v1/chat/completions:
Embedded-jinja path (llama.cpp, llama-cpp-python, LM Studio). The Qwen 3.6 native chat template baked into the GGUF instructs the model to emit a more verbose XML form. This is the shape you'll see if you talk to llama-server or LM Studio directly:
Pick the parser shape that matches your loader. Don't mix.
Example (Ollama, OpenAI-compatible API)
python
1from openai import OpenAI
23client = OpenAI(base_url="http://localhost:11434/v1", api_key="ignored")45resp = client.chat.completions.create(6 model="janus",7 messages=[8{"role":"user","content":"Call get_weather for Tokyo. Respond ONLY with the tool call."}9],10 tools=[{11"type":"function",12"function":{13"name":"get_weather",14"description":"Get current weather for a city",15"parameters":{16"type":"object",17"properties":{"city":{"type":"string"}},18"required":["city"],19},20},21}],22 temperature=0.3,23)24print(resp.choices[0].message.tool_calls)25# [ToolCall(id='call_xxx', type='function',26# function=Function(name='get_weather', arguments='{"city":"Tokyo"}'))]
Tips
Use direct prompts ("Call X for Y") rather than soft hints ("Use the tool"). The model thinks before committing to a call, and weak prompts can exhaust num_predict inside the <think> block before the call is emitted.
Allow at least num_predict: 1024 (or max_tokens: 1024) for tool-calling turns, more if the schemas are large.
The Modelfile's JSON-in-XML format is what Ollama's tool-call extractor understands; if you swap loaders, swap the parser to match (see "Embedded-jinja path" above).
Known limitations
No mmproj in this release. The base Qwen3.6 supports image and video input via a separate mmproj file, which is not included here. Text-only inference works out of the box; multimodal inference requires fetching Qwen3.6-35B-A3B-uncensored-heretic-mmproj-BF16.gguf (or equivalent) from upstream — run ./scripts/fetch_vision.sh and see Vision for the full path.
Quantization-induced quality loss. Q4_K_M is a strong general-purpose quant but does measurably degrade math and code accuracy compared to BF16. If you need maximum quality, run the upstream safetensors on a GPU that fits BF16 (~70 GB).
MoE expert utilization is uneven. Stock Qwen3.6-35B-A3B routes 8 of 256 experts per token. On narrow domains (e.g. only one programming language) a small subset of experts dominates; load-balance loss was a training-time concern, not a runtime guarantee.
Thinking traces can loop. Like most reasoning-distilled models, Janus-35B occasionally gets stuck repeating itself inside <think> tags. Mitigations: lower temperature to 0.4-0.6, raise repeat_penalty to 1.08, or set a <think>-token budget cap if your loader supports it.
Large tool-call arguments can be dropped. Ollama's JSON-in-XML tool format makes the model JSON-escape the entire arguments object inline; for a big/complex payload (e.g. a file's content in a write_file call) the model can fail to escape it, so the field arrives undefined and the call fails. Qwen's native <function=…><parameter=…> format (raw values, no escaping) was tested as a fix but parses unreliably through Ollama, so the template deliberately keeps JSON-in-XML. Mitigation: write large files in smaller pieces per call.
Uncensored base — not aligned with any specific safety policy. This is a personal repackage of an open-weight base whose refusal behavior has been abliterated away (the llmfan46 Heretic base). There is no RLHF refusal layer; the model will attempt most requests, so downstream safety is entirely the operator's responsibility.
No formal evaluation in this card. Most numbers in the hardware table are estimates; the Z13 row (28.71 tok/s at Q4_K_M) is measured. If you produce real benchmarks (MMLU, HumanEval, etc.) and want them included, file a PR.
Dense sibling, rebased to the Qwen 3.8heretic-org/Qwen3.8-27B-heretic-ara base. Same teacher (Fable 5), same dataset family, smaller memory footprint, no MoE quirks. (The older FoolDev/Thanatos-27B and Thanatos-27B-Heretic slugs now 307 to this path.)
Heretic-flavored fine-tune on a smaller 9B Qwen base. Useful as a fast first-pass model when 35B is too heavy for the host.
Why Janus stays on Qwen 3.6: the 35B-A3B MoE config was discontinued in Qwen 3.8 — the 3.8 MoEs are Qwen3.8-Flash-Next (~180B) and Qwen3.8-2.4T-A95B (2.4T), neither anywhere near this scale, and llmfan46 published no 3.8 heretic. So Janus stays on the 3.6 base (the newest 35B-A3B heretic that exists), while the dense sibling Thanatos-27B moved to Qwen 3.8 (whose dense model is 27B).
