Table with columns: Tensors, Format| Tensors | Format |
|---|
| attention q, k, v, o (8 layers) | bfloat8_b, 8.5 bits a weight |
| Gated DeltaNet q, k, v, z, a, b, output (24 layers) | bfloat8_b, 8.5 bits a weight |
| MLP gate, up, down (32 layers) | bfloat8_b, 8.5 bits a weight |
| LM head (tied to the embedding), embedding, norms and every other tensor | bf16, except the original dtypes below |
mtp, visual | bf16, unchanged |
Original tensors: model.language_model.layers.N.linear_attn.A_log, model.language_model.layers.N.linear_attn.norm.weight | float32, unchanged |
9.63 bits a weight on average, 5.07 GB packed. quant4.json lists each projection's format. Quantized with
GPTQ on the bfloat8_b grid, keeping each block's shared exponent as the packer sets it, calibrated on 65,536
tokens of WikiText-2 and Korean Wikipedia.
Only the language model was quantized and evaluated.
Quickstart
from transformers import AutoModelForCausalLM, AutoTokenizer repo = "Xylolabs/Qwen3.5-4B-Tenstorrent-Tensix-BFP8"tok = AutoTokenizer.from_pretrained(repo)model = AutoModelForCausalLM.from_pretrained(repo, dtype="bfloat16", device_map="auto") messages = [{"role": "user", "content": "What is the capital of France?"}]inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)out = model.generate(**inputs, max_new_tokens=64)print(tok.decode(out[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))
Run on Tensix
Checked with ttnn 0.81.1 on ttsim, Tenstorrent's Blackhole simulator. Pack one matrix and
check that no value changed (TT-NN reads weights as [in, out]):
import jsonimport torchimport ttnnfrom safetensors import safe_open d = "Qwen3.5-4B-Tenstorrent-Tensix-BFP8" # hf download Xylolabs/Qwen3.5-4B-Tenstorrent-Tensix-BFP8 --local-dir Qwen3.5-4B-Tenstorrent-Tensix-BFP8name = "model.language_model.layers.0.mlp.down_proj.weight"fmt = json.load(open(f"{d}/quant4.json"))["layers"]["0"]["mlp.down_proj"]["format"]dtype = {"bfp4": ttnn.bfloat4_b, "bfp8": ttnn.bfloat8_b}[fmt]shard = json.load(open(f"{d}/model.safetensors.index.json"))["weight_map"][name]with safe_open(f"{d}/{shard}", "pt") as f: w = f.get_tensor(name).t().contiguous() device = ttnn.open_device(device_id=0)w_tt = ttnn.as_tensor(w, dtype=dtype, layout=ttnn.TILE_LAYOUT, device=device, memory_config=ttnn.DRAM_MEMORY_CONFIG)assert torch.equal(ttnn.to_torch(w_tt).float(), w.float())ttnn.close_device(device)
python pack_tensix.py Qwen3.5-4B-Tenstorrent-Tensix-BFP8 checks every matrix this way (--device also round-trips it
through device DRAM). python verify_grid.py Qwen3.5-4B-Tenstorrent-Tensix-BFP8 checks the grid with torch alone.
tt-metal Qwen3.5/3.6 demo (v0.81.1). tt-transformers does not support this architecture. tt-metal's
Qwen3.5/3.6 demo does
not list Qwen3.5-4B but reads every shape from config.json, and it does not run this checkpoint at its
precision. It always packs the MLP gate and up projections in bfloat4_b, which re-quantizes these bfloat8_b
values. Its DeltaNet, attention and MLP down projections packed exactly in bfloat8_b (checked on layers 0 to
3). The LM head and the MTP block, kept in bf16 here, are also re-quantized to bfloat8_b. For this demo, the
BFP4 export matches what it runs.
tt-forge (tt-xla). Not run with this checkpoint yet. By default it keeps the weights in bf16, which is
exact. Its bfp_bf8 option re-quantizes them in PyTorch's [out, in] layout.
Evaluation
Table with columns: Bits a weight, Size (GB), Mean KL to BF16, Top-1 (%), Answers right, Lost | Bits a weight | Size (GB) | Mean KL to BF16 | Top-1 (%) | Answers right | Lost |
|---|
| BF16 (Qwen/Qwen3.5-4B) | 16 | 8.41 | 0 | 100 | 36 of 40 | - |
| this model | 9.63 | 5.065 | 0.00084 | 98.62 | 35 of 40 | 0 |
| Q4_K_M (llama.cpp) | 5.13 | 2.783 | 0.03981 | 89.79 | 37 of 40 | 0 |
KL and top-1 cover 16,312 positions of English and Korean text (WikiText-2 test, Korean Wikipedia), plain and
in the chat template. Answers are 33 short prompts with automatic checks, plus the 7 reasoning ones again with
thinking on (40 in all). "Lost" means right in BF16 but wrong or broken here; near ties are not counted.
Q4_K_M was made from the same checkpoint with llama.cpp v0.9.0 (cd959bae) and no importance matrix; its
sha256 is 9bb84287c48e7d3b5e0b04f4809c62e7ec754e373b6a7a0bf3e737170b709334. Our bar: mean KL at most 1.5
times Q4_K_M's and at most one more lost answer. Passed: 0.02 times Q4_K_M's KL, 0 lost against 0.
These numbers measure closeness to BF16, not task quality. Speed is not reported here.
License
Apache 2.0, as the base model (LICENSE). The changes are listed in NOTICE. pack_tensix.py and
verify_grid.py are by Xylolabs under the same license. Provided as is, without warranty. Not affiliated with
or endorsed by Tenstorrent.
Citation
@misc{qwen3.5, title = {{Qwen3.5}: Towards Native Multimodal Agents}, author = {{Qwen Team}}, month = {February}, year = {2026}, url = {https://qwen.ai/blog?id=qwen3.5}}