ornith-1.0-35b-p150 on Tenstorrent p150 and p150x2 and p150x4

Serves ornith-ai/Ornith-1.0-35B by Ornith AI at revision 5df2ed3f. See the original model card for training and evaluation details.

Ornith-1.0-35B, a 35B-parameter mixture-of-experts model for agentic reasoning and coding, served with vLLM with tool calling and separated reasoning output.

Runs on Tenstorrent p150 or p150x2 or p150x4 or p100 — see the serving profiles below.

Packaged and published with tt-model-manager 0.1.0 (manifest schema 5.1).

At a glance

Hardware	p150, p150x2, p150x4
Context	262,144 tokens
License	mit

Intended use

Agentic coding and reasoning through an OpenAI-compatible API, for example from OpenCode.

Prerequisites

• The Tenstorrent CLI, tt: uv tool install tenstorrent (or use tt-model alone, see the Quickstart)
• Docker
• Tenstorrent hardware: p150 or p150x2 or p150x4

Quickstart

tt serve tt-hous/ornith-1.0-35b-p150

tt serve downloads the Docker image and the ornith-ai/Ornith-1.0-35B at 5df2ed3f675c7beaa490328cc70bb573b65fb660 weights (into your HF cache; they are not in the image), then starts an OpenAI-compatible server on port 20000 (or the next free port, if that one is busy). The first start compiles kernels for your device, which takes several minutes; the server is ready when it logs Server ready after.

Without tt-cli: tt-model serve tt-hous/ornith-1.0-35b-p150.

To use it from OpenCode, add an OpenAI-compatible provider with the base URL the serve command printed and the model ornith-ai/Ornith-1.0-35B.

Serving profiles

Select a different serving profile using --profile.

profile	hardware	mesh	max_num_seqs	max_model_len	description
p150	p150	P150	1	262144	One interactive user on one P150 board.
p100	p100	P150	1	262144	One p100, same settings as p150: 262144 context, 1 sequence.
p150x2	p150x2	P150x2	1	262144	One interactive user on both P150-class ASICs of one P300C board using tensor and expert parallelism.
p150x4 (default)	p150x4	P150x4	1	262144	One interactive user on all four P150-class ASICs across two P300C boards using tensor and expert parallelism.

Capabilities

The server speaks the OpenAI API at http://127.0.0.1:20000/v1 (or whichever port your serve command reported; chat completions, completions, and /v1/models). Pass "model": "ornith-ai/Ornith-1.0-35B" — the weights id, not this package's name.

Tool calling is enabled (--tool-call-parser qwen3_xml): a request that passes tools comes back with tool_calls and finish_reason: tool_calls.

Reasoning output is separated (--reasoning_parser qwen3): the thinking text arrives in reasoning_content, apart from content.

The model card's sampling for general tasks is temperature 0.6, top_p 0.95, top_k 20.

Expected performance

eval	this package	reference	prompts	notes
GPQA Diamond, exact match	67.5%	—	40 of 198	13 of 40 responses hit the 4,096-token cap; standard error 7.5
LiveCodeBench v6, 2025 Jan to Apr problems	48.0%	—	50 of 175	22 of 50 responses hit the 4,096-token cap; standard error 7.1
IFEval, prompt-level strict / loose	79.0% / 83.0%	—	100 of 541	2,048-token cap; standard error 4.1 (strict)

Evals ran on 2026 Sep 30 on p150x4, against this package's image, with lm-evaluation-harness through the chat API: temperature 0.6, top_p 0.95, top_k 20, thinking off, 0-shot, one request at a time. The original model card reports only agentic-coding benchmarks (SWE-bench), so there is no reference figure for these evals.

profile	ISL	OSL	concurrency	N	TTFT (ms)	prefill tok/s/u	decode tok/s/u	E2EL (s)
p150	128	128	1	1	253	506	11.6	11.21
p150	2,048	128	1	1	2,601	787	11.6	13.58
p150	8,192	128	1	1	10,409	787	11.5	21.42
p150	32,768	128	1	1	42,786	766	11.4	53.91
p150x2	128	128	1	1	153	837	33.3	3.97
p150x2	2,048	128	1	1	1,404	1,459	33.2	5.23
p150x2	8,192	128	1	1	5,912	1,386	32.9	9.77
p150x2	32,768	128	1	1	24,468	1,339	32.0	28.44
p150x4	128	128	1	1	160	800	43.0	3.11
p150x4	2,048	128	1	1	801	2,557	42.7	3.78
p150x4	8,192	128	1	1	3,475	2,357	42.3	6.47
p150x4	32,768	128	1	1	14,450	2,268	40.8	17.57
p150x4	65,536	128	1	1	29,950	2,188	38.8	33.22
p150x4	131,072	128	1	1	65,211	2,010	35.8	68.76
p150x4	262,016	128	1	1	149,229	1,756	30.4	153.40

Medians of three sequential requests per row, measured with vllm bench serve on one QuietBox 2 (two P300C boards): one ASIC for p150, both ASICs of one board for p150x2, all four for p150x4. Random prompts of exactly ISL tokens, output pinned to OSL tokens, greedy. Prefill tok/s/u is ISL divided by TTFT, decode tok/s/u is 1,000 divided by TPOT (ms), and E2EL is TTFT plus 127 times TPOT. The last p150x4 row fills the full 262,144-token context.

Limitations

• One request at a time on every profile. Tested on one QuietBox 2 only.
• Accuracy was measured on subsets with thinking off, on p150x4 only. With a 4,096-token cap, a third of GPQA and nearly half of LiveCodeBench responses ran out of budget before answering. No agentic-coding benchmark was run.
• Not for interactive use on p150: decode is about 11.5 tokens per second, and a 32,768-token prompt takes about 43 s to first token.

Feedback

Questions or problems with this package: open a discussion at https://huggingface.co/tt-hous/ornith-1.0-35b-p150/discussions — that is what reaches its author. A problem with the tt tooling itself: tt report issue (collects your environment and opens a prefilled issue against tenstorrent/tt-cli). Product feedback: support@tenstorrent.com.

Provenance

The exact sources the image was built from — code/ in this repo is byte-identical to the model code inside the image:

component	built from
tt-metal	90bc8b5dba353464927ea4df0944b5bcab79c6a0
vLLM	a887998646dc4e6f192bce8d485bf89f4596ca2f
code/ digest	fe60775587d80f7d (sha256, first 16 hex digits)
built	2026-09-14T14:57:23+00:00 by tt-model 0.1.0