keptn/ornith-1.5-9b-bf16

Ornith 1.5 9B · bf16 · for radiance

ornith-ai/Ornith-1.5-9B in one 16.6 GiB file for the radiance inference engine on AMD RDNA4. A dense 9B reasoning model, tuned for coding and agent work. It runs on one card.

File	ornith-1.5-9b-bf16.rad, 16.59 GiB
Needs	radiance 1.15.1 or newer; 1 card with 32 GB (tested on Radeon AI PRO R9700), or 2 cards with --tp 2
Format	weights bf16 as in the original; lm_head FP8, 128×128 blocks; activations bf16
Speculation	none (radiance's bf16 plugin serves the trunk only; the MTP head is not in this file)
Inputs	text · image
Context	262,144 tokens
Version	1.0 (2026-11-03)

Speed

Radeon AI PRO R9700 at 215 W, radiance 1.15.1, fp8 KV, no draft tokens, thinking off. BetterBench 0.6.1, full run, temperature 0.7. Tokens per second (median):

	1 card	2 cards
decode, one stream	32.1	62.4
prefill at 24k tokens	5,153	8,360
4 streams, total	110	215

Quality

Distance to the bf16 original on the 80 MT-Bench questions, both turns. The text is the GPT-4 answers from lmsys/mt_bench_human_judgments, with the model's chat template and thinking off. The reference is the bf16 original in HF transformers, with fp32 arithmetic and the full vocabulary. radiance scores this file with --kld-ref, on the answer tokens only (42,305 tokens).

	mean KL	99th percentile KL	top token the same	perplexity ratio
this file	0.00086	0.00753	98.9%	×1.001
keptn/ornith-1.5-9b-fp8	0.00498	0.0514	97.8%	×1.001

Quick start

With the compose files from the radiance repository (1.15.1 or newer).

Or directly (one card):

hf download keptn/ornith-1.5-9b-bf16 ornith-1.5-9b-bf16.rad --local-dir ~/models
docker run -d --init --name ornith --network host --group-add video --group-add render \
    --device /dev/kfd --device /dev/dri --security-opt seccomp=unconfined --ulimit memlock=-1:-1 \
    -v ~/models:/models:ro -e ROCR_VISIBLE_DEVICES=0 \
    stilldeadcode/radiance:1.15.1 --model /models/ornith-1.5-9b-bf16.rad --host 0.0.0.0 --port 8000 \
    --kv-cache-dtype fp8 --max-model-len 262144 --max-num-seqs 4 \
    --max-num-batched-tokens 2048 \
    --temp 1.0 --top-p 0.95 --top-k 20 --presence-penalty 1.5

	change
two cards	ROCR_VISIBLE_DEVICES=0,1, add --tp 2 --p2p on
coding	--temp 0.6 --presence-penalty 0 (Ornith's setting for precise coding)
thinking	on by default; per request "chat_template_kwargs": {"enable_thinking": false} turns it off

Memory and context

KV pool at --max-model-len 262144, fp8 KV, radiance 1.15.1, from radiance's startup report. Tokens of context in total, shared by all sequences:

	2 sequences	4 sequences	8 sequences
1 card	524,320	1,048,592	1,042,608
2 cards	524,384	1,048,656	2,097,200

Images, audio and video

The same inputs go to radiance and to the bf16 original in HF transformers. Greedy, thinking off, 1 card, radiance 1.15.1 (release image). There is one measurement for each vision or audio encoder, made with ornith-1.5-9b-fp8.rad.

input	cases	same answer	mean KL	top token the same
images	20	10/20	0.102	244/264

Not measured on this file. The image test ran on the FP8 file. Both files keep the same bf16 vision tower.

How it was made

1. Start from the bf16 release ornith-ai/Ornith-1.5-9B (revision 489cb979).
2. Keep every weight as it is.
3. Round only the lm_head to FP8, with one scale per 128×128 block. This follows radiance's Qwen3.8-27B bf16 recipe.

rad-convert Ornith-1.5-9B --tokenizer Ornith-1.5-9B/tokenizer.json \
    --recipe ornith-1.5-9b-bf16.recipe -o ornith-1.5-9b-bf16.rad

Calibration: no calibration data.

Model card with: radiance-model-cards-framework 0.3 (codeberg.org/keptn42/radiance-model-cards-framework).

Limits

• radiance runs on AMD RDNA4 cards.
• This file has no speculation. The bf16 plugin does not serve the MTP head. For speed, use the FP8 file.
• With thinking on, hard tasks can need more than 32k output tokens.
• Speed on four cards is not measured.

Credits

• Model: the Ornith team (github.com/ornith-ai/Ornith-1), built on Qwen3.5 (Apache 2.0).
• radiance: StillDeadcode (codeberg.org/StillDeadcode/radiance).
• Test sets: MT-Bench, DocVQA, AI2D.

License

MIT, the license of the base model. See LICENSE (from github.com/ornith-ai/Ornith-1).