A multimodal fine-tune of Qwen 3.6 35B-A3B, the latest-generation Qwen Mixture-of-Experts base model (35B total parameters, ~3B active per token). Part of the Ornstein series — reasoning- and agent-oriented fine-tunes built on a custom data curation pipeline.

GGUF quantizations available at DJLougen/Ornstein3.6-35B-A3B-GGUF — Q8_0 down through aggressive 2-bit I-quants, all with imatrix calibration.

Support This Work

I'm a PhD student in visual neuroscience at the University of Toronto who also happens to spend way too much time fine-tuning, merging, and quantizing open-weight models on rented H100s and a local DGX Spark. All training compute is self-funded — balancing GPU costs against a student budget. If my uploads have been useful to you, consider buying a PhD student a coffee. It goes a long way toward keeping these experiments running.

Support on Ko-fi

Vision/video restoration: The repository now includes the Qwen3.6 base visual tower plus image/video processor files, so the source safetensors checkpoint should load as the multimodal conditional-generation architecture. Existing GGUF artifacts made before this restoration remain text-only until rebuilt.

Details

Developed by: DJLougen
Architecture: Qwen3_5MoeForConditionalGeneration — Qwen 3.6 MoE with linear + full attention interleaved (Gated Delta Net)
Parameters: 34.66B total, ~3B active (256 experts, 8 active per token)
Hidden size / layers: 2048 / 40
Context length: 262,144 tokens
License: Apache 2.0
Base model: unsloth/Qwen3.6-35B-A3B
Training framework: Unsloth

Usage

Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "DJLougen/Ornstein3.6-35B-A3B"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")

messages = [{"role": "user", "content": "Explain mixture-of-experts routing in one paragraph."}]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=512)
print(tok.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

llama.cpp (via GGUF)

See the GGUF repo — pick a quant that fits your memory (Q4_K_M is a strong default for 24GB).

License

Apache 2.0 — inherited from the Qwen 3.6 base release.

GestaltLabs/Ornstein3.6-35B-A3B

Ornstein3.6-35B-A3B