Releases: tenstorrent/tt-inference-server
Releases · tenstorrent/tt-inference-server
Release list
v0.20.0
Summary of Changes
- Uplifted Qwen3.6-27B on on Blackhole QuietBox 2
- Added support for Qwen3-Embedding-4B on Blackhole QuietBox 2
- Added support for Qwen3-Embedding-0.6B on Blackhole QuietBox 2
- Added support for bge-m3 on Blackhole QuietBox 2
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
qwen36_blackhole |
Qwen3.6-27B |
Qwen/Qwen3.6-27B |
P300X2 | de59f8a |
No change [EXPERIMENTAL] | CI Link |
forge_vllm_plugin |
Qwen3-Embedding-4B |
Qwen/Qwen3-Embedding-4B |
P300X2 | de59f8a |
[EXPERIMENTAL] | CI Link |
forge_vllm_plugin |
Qwen3-Embedding-0.6B |
Qwen/Qwen3-Embedding-0.6B |
P300X2 | de59f8a |
[EXPERIMENTAL] | CI Link |
forge_vllm_plugin |
bge-m3 |
BAAI/bge-m3 |
P300X2 | de59f8a |
[EXPERIMENTAL] | CI Link |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.20.0-de59f8a-03fa3af
- https://ghcr.io/tenstorrent/tt-media-inference-server-forge:0.20.0-de59f8a
Total: 2
v0.19.0
Summary of Changes
- Uplifted Llama-3.1-8B-Instruct on p300
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
tt_transformers |
Llama-3.1-8B-Instruct |
meta-llama/Llama-3.1-8B-Instruct |
P300 | 8c48a10 → b204341 |
No change [FUNCTIONAL] | CI Link |
Release Artifacts Summary
Images Promoted from Models CI
Total: 1
v0.18.0
Summary of Changes
- Uplifted Llama-3.1-8B-Instruct on p150
- Added support for Qwen3.6-27B on Blackhole QuietBox 2
- Added support for gemma-4-31b-it on Blackhole QuietBox 2
- Added support for Falcon3-7B-Instruct on p150
- Added support for yolox_nano on p150
- Uplifted FLUX.1-schnell on Blackhole QuietBox 2
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
tt_transformers |
Llama-3.1-8B-Instruct |
meta-llama/Llama-3.1-8B-Instruct |
P150 | 55fd115 → c49bb76 |
No change [EXPERIMENTAL] | CI Link |
qwen36_blackhole |
Qwen3.6-27B |
Qwen/Qwen3.6-27B |
P300X2 | c49bb76 |
[EXPERIMENTAL] | CI Link |
tt_transformers |
gemma-4-31B-it |
google/gemma-4-31B-it |
P300X2 | c49bb76 |
[EXPERIMENTAL] | CI Link |
forge_vllm_plugin |
Falcon3-7B-Instruct |
tiiuae/Falcon3-7B-Instruct |
P150 | c49bb76 |
[EXPERIMENTAL] | CI Link |
tt_transformers |
yolox_nano |
yolox_nano |
P150 | c49bb76 |
[COMPLETE] | CI Link |
tt_transformers |
FLUX.1-schnell |
black-forest-labs/FLUX.1-schnell |
P300X2 | 55fd115 → c49bb76 |
No change [COMPLETE] | CI Link |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.18.0-c49bb76-6b4a3a7
- https://ghcr.io/tenstorrent/tt-media-inference-server:0.18.0-c49bb76
- https://ghcr.io/tenstorrent/tt-media-inference-server-forge:0.18.0-c49bb76
Total: 3
v0.17.0
Summary of Changes
- Uplifted gpt-oss-120b on WH Galaxy
- Added support for gpt-oss-120b on Blackhole QuietBox 2
- Uplifted Llama-3.1-8B-Instruct on Blackhole QuietBox 2
- Uplifted Qwen3-32B on Blackhole QuietBox 2
- Added support for Z-Image-Turbo on Blackhole QuietBox 2
- Uplifted Wan2.2-T2V-A14B-Diffusers on Blackhole QuietBox 2
- Uplifted FLUX.1-dev on Blackhole QuietBox 2
- Uplifted whisper-large-v3 on WH Galaxy
- Uplifted distil-large-v3 on WH Galaxy
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.8.1
- tt-kmd: 2.7.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
gpt_oss |
gpt-oss-120b |
openai/gpt-oss-120b |
GALAXY | 805f43d → 8c48a10 |
No change [EXPERIMENTAL] | CI Link |
gpt_oss |
gpt-oss-120b |
openai/gpt-oss-120b |
P300X2 | 805f43d → 8c48a10 |
EXPERIMENTAL | CI Link |
tt_transformers |
Llama-3.1-8B-Instruct |
meta-llama/Llama-3.1-8B-Instruct |
P300X2 | 80180b9 → 8c48a10 |
No change [FUNCTIONAL] | CI Link |
qwen3_32b_galaxy_impl |
Qwen3-32B |
Qwen/Qwen3-32B |
P300X2 | 80180b9 → 8c48a10 |
No change [FUNCTIONAL] | CI Link |
tt_transformers |
Z-Image-Turbo |
Tongyi-MAI/Z-Image-Turbo |
P300X2 | 8c48a10 |
TOP PERF | CI Link |
tt_transformers |
Wan2.2-T2V-A14B-Diffusers |
Wan-AI/Wan2.2-T2V-A14B-Diffusers |
P300X2 | 80180b9 → 8c48a10 |
No change [COMPLETE] | CI Link |
tt_transformers |
FLUX.1-dev |
black-forest-labs/FLUX.1-dev |
P300X2 | 80180b9 → 8c48a10 |
No change [COMPLETE] | CI Link |
whisper |
whisper-large-v3 |
openai/whisper-large-v3 distil-whisper/distil-large-v3 |
GALAXY | 25891d3 → 8c48a10 |
No change [COMPLETE] | CI Link |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.17.0-8c48a10-f52987a
- http://ghcr.io/tenstorrent/tt-media-inference-server:0.17.0-8c48a10
Total: 2
v0.16.0
Summary of Changes
- Uplifted Llama-3.3-70B-Instruct on Blackhole QuietBox 2
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.2.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
tt_transformers |
Llama-3.3-70B-Instruct |
meta-llama/Llama-3.3-70B-Instruct |
P300X2 | 555f240 → 669d59e |
No change [FUNCTIONAL] | CI Link |
Release Artifacts Summary
Images Promoted from Models CI
Total: 1
v0.15.0
Summary of Changes
- Introduced first tt-inference-server Helm chart https://github.com/tenstorrent/tt-inference-server/blob/main/charts/tt-inference-server/README.md
- supports all tt-media-server based ModelSpecs
- supports all vLLM based ModelSpecs released after v0.11.0
- Uplifted speecht5_tts on p150 and Blackhole QuietBox 2
- Uplifted whisper-large-v3 on p150 and Blackhole QuietBox 2
- Uplifted distil-large-v3 on p150 and Blackhole QuietBox 2
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.2.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
speecht5-tts |
speecht5_tts |
microsoft/speecht5_tts |
P150, P300X2 | 2508216 → 25891d3 |
No change [EXPERIMENTAL] | N/A |
whisper |
whisper-large-v3 |
openai/whisper-large-v3 distil-whisper/distil-large-v3 |
P150, P300X2 | 2508216 → 25891d3 |
No change [COMPLETE] | N/A |
Release Artifacts Summary
Images Promoted from Models CI
Total: 1
v0.14.0
Summary of Changes
- Addressed #2518
run.pyCLI now clearly indicates which release of tt-inference-server to use for LLM and VLM models that have not been uplifted since v0.10.0 or older
- Updated model support table to include Blackhole QuietBox 2 models (p300x2), see Models by Hardware page
- Added missing Blackhole QuietBox 2 model pages in
docs/model_support/ - Uplifted FLUX.1-dev on Blackhole QuietBox 2
- Uplifted Wan2.2-T2V-A14B-Diffusers on Blackhole QuietBox 2
- Uplifted Llama-3.1-8B-Instruct on Blackhole QuietBox 2
- Uplifted Qwen3-32B on Blackhole QuietBox 2
- Uplifted Qwen3-VL-32B-Instruct on WH T3K
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.2.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
tt_transformers |
Qwen3-VL-32B-Instruct |
Qwen/Qwen3-VL-32B-Instruct |
T3K | 079a2c2 → 80180b9 |
EXPERIMENTAL → FUNCTIONAL | N/A |
tt_transformers |
FLUX.1-dev |
black-forest-labs/FLUX.1-dev |
P300X2 | 079a2c2 → 80180b9 |
No change [COMPLETE] | N/A |
tt_transformers |
Wan2.2-T2V-A14B-Diffusers |
Wan-AI/Wan2.2-T2V-A14B-Diffusers |
P300X2 | 079a2c2 → 80180b9 |
No change [COMPLETE] | N/A |
qwen3_32b_galaxy_impl |
Qwen3-32B |
Qwen/Qwen3-32B |
P300X2 | 079a2c2 → 80180b9 |
No change [FUNCTIONAL] | N/A |
tt_transformers |
Llama-3.1-8B-Instruct |
meta-llama/Llama-3.1-8B-Instruct |
P300X2 | 079a2c2 → 80180b9 |
No change [FUNCTIONAL] | N/A |
Release Artifacts Summary
Images Promoted from Models CI
- http://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.14.0-80180b9-7678b70
- http://ghcr.io/tenstorrent/tt-media-inference-server:0.14.0-80180b9
Total: 2
v0.13.0
Summary of Changes
- Initial release of 5 Forge models, with the newest version of the forge library (1.0.0)
SW versions recommended for Wormhole Galaxy:
- tt-smi: 3.0.38
- Firmware: 19.3.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
tt_transformers |
resnet-50 |
resnet-50 |
N150, N300 | 2496be4 → 079a2c2 |
EXPERIMENTAL → FUNCTIONAL | N/A |
tt_transformers |
vovnet |
vovnet |
N150, N300 | 2496be4 → 079a2c2 |
EXPERIMENTAL → COMPLETE | N/A |
tt_transformers |
mobilenetv2 |
mobilenetv2 |
N150, N300 | 2496be4 → 079a2c2 |
EXPERIMENTAL → COMPLETE | N/A |
tt_transformers |
segformer |
segformer |
N150, N300 | 2496be4 → 079a2c2 |
EXPERIMENTAL → FUNCTIONAL | N/A |
tt_transformers |
vit |
vit |
N150, N300 | 2496be4 → 079a2c2 |
EXPERIMENTAL → COMPLETE | N/A |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-media-inference-server-forge:0.13.0-079a2c2 from https://ghcr.io/tenstorrent/tt-shield/tt-media-inference-server-forge:079a2c23f4b360dd0c415a43dd2ffc94d0a792de_fbccfcd_72718973516
Total: 1
v0.10.1
Summary of Changes
- Fix a deadlock in the media server
- Increase the trace region size for Llama-3.1-70B models
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.2.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit |
|---|---|---|---|---|
tt_transformers |
Qwen3-32B |
Qwen/Qwen3-32B |
P150X8, P300X2 | 555f240 |
tt_transformers |
Llama-3.3-70B-Instruct |
meta-llama/Llama-3.3-70B-Instructmeta-llama/Llama-3.1-70Bmeta-llama/Llama-3.1-70B-Instructdeepseek-ai/DeepSeek-R1-Distill-Llama-70B |
P150X4, P150X8, P300X2 | 555f240 |
tt_transformers |
Llama-3.1-8B |
meta-llama/Llama-3.1-8Bmeta-llama/Llama-3.1-8B-Instruct |
P150X8, P300X2, P300 | 555f240 |
tt_transformers |
Wan2.2-T2V-A14B |
Wan-AI/Wan2.2-T2V-A14B-Diffusers |
P150X8, P300X2 | 555f240 |
tt_transformers |
FLUX.1-dev |
black-forest-labs/FLUX.1-devblack-forest-labs/FLUX.1-schnell |
P300X2, P150X8 | 555f240 |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.10.1-555f240-22be241
- https://ghcr.io/tenstorrent/tt-media-inference-server:0.10.1-555f240
Total: 2
v0.12.0
Summary of Changes
- Updated multi-host model documentation pages, see WH Dual-Galaxy deepseek-ai/DeepSeek-R1-0528
- Uplift deepseek-ai/DeepSeek-R1-0528
- Still in experimental mode as the model hangs during decode
- Uplift openai/gpt-oss-120b
- Still in experimental mode as the model sometimes returns incomplete special tokens
SW versions recommended for Wormhole Galaxy:
- tt-smi: 4.0.0
- Firmware: 19.2.0
- tt-kmd: 2.5.0
Model Spec Release Updates
This document shows model specification updates.
| Impl | Model Arch | Weights | Devices | TT-Metal Commit Change | Status Change | CI Job Link |
|---|---|---|---|---|---|---|
gpt-oss |
gpt-oss-120b |
openai/gpt-oss-120b |
GALAXY, T3K | bac8b34 → 805f43d |
No change | N/A |
deepseek_r1_galaxy |
DeepSeek-R1-0528 |
deepseek-ai/DeepSeek-R1-0528 |
GALAXY, DUAL_GALAXY, QUAD_GALAXY | bac8b34 → 805f43d |
No change | N/A |
Release Artifacts Summary
Images Promoted from Models CI
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-release-ubuntu-22.04-amd64:0.12.0-805f43d-a45c614
- https://ghcr.io/tenstorrent/tt-inference-server/vllm-tt-metal-src-multihost-ubuntu-22.04-amd64:0.12.0-805f43d-a45c614
Total: 2