开发者

在Tenstorrent硬件上快速运行您的模型。通过两个开源SDK，您可以尽可能接近底层硬件，或让我们的AI编译器来完成工作。

模型

Explore models optimized on Tenstorrent hardware.

Don’t see yours listed? Check out TT-Forge, our compiler, to get other models running today.

模型46
bge-large-en-v1.5

Purpose-built for semantic search, dense retrieval, and RAG. BAAI, 335M.

Feature Extraction
335M
BGE-M3

Multi-lingual, multi-granularity retrieval across dense, sparse, and multi-vector modes. BAAI, 568M.

Feature Extraction
568M
DeepSeek-R1

Mixture-of-experts reasoning powerhouse rivaling closed frontier models on math and code. DeepSeek, 671B.

Text Generation
671B
distil-large-v3

Distilled Whisper large-v3 — near-parity transcription at roughly six times the speed. Distil-Whisper, 756M.

Speech-to-Text
756M
Falcon-40B-Instruct

Commercially licensed generalist trained on 1T tokens of curated web data. TII, 40B.

Text Generation
40B
Falcon-7B-Instruct

Commercially licensed instruction model — lean resource footprint. TII, 7B.

Text Generation
7B
FLUX.1 [dev]

Flow-matching transformer for photorealistic text-to-image with strong prompt adherence. Black Forest Labs, 12B.

Text-to-Image
12B
FLUX.1 [schnell]

FLUX distilled to 4 steps — full quality, fraction of the compute. Black Forest Labs, 12B.

Text-to-Image
12B
gemma-3-27b

Largest Gemma 3 base — 140-language reasoning, coding, long context. Google, 27B.

Text Generation
27B
gemma-3-4b-it

Compact Gemma 3 base — punches above its weight in reasoning and code. Google, 4B.

Text Generation
4B
gemma-3-4b-it

Punches above its weight in reasoning and code. Google, 4B.

Text Generation
4B
gpt-oss-120b

Open GPT-style model for deep reasoning in self-hosted deployments. GPT-OSS, 120B.

Text Generation
120B
gpt-oss-20b

Open GPT-style generation at a practical self-hosted scale. GPT-OSS, 20B.

Text Generation
20B
Llama-3.1-8B

128K context and multilingual base with a wide fine-tuning ecosystem. Meta, 8B.

Text Generation
8B
Llama-3.1-8B-Instruct

Multilingual instruction following, tool use, and function calling. Meta, 8B.

Text Generation
8B
Llama-3.2-11B-Vision

Image + text inputs on Llama's reasoning foundation, 128K context. Meta, 11B.

Image-Text-to-Text
11B
Llama-3.2-1B-Instruct

Instruction-tuned to run anywhere — on-device at 1B. Meta, 1B.

Text Generation
1B
Llama-3.2-3B-Instruct

Lightweight agent backbone for low-latency, resource-constrained deployments. Meta, 3B.

Text Generation
3B
Llama-3.2-90B-Vision

Image + text inputs at full scale on Llama's reasoning foundation. Meta, 90B.

Image-Text-to-Text
90B
Llama-3.3-70B

Llama 3.1 refined — better math, code, and multilingual, same 128K context. Meta, 70B.

Text Generation
70B
Llama-3.3-70B-Instruct

Stronger structured tasks, tool use, and reasoning than prior Llama generations. Meta, 70B.

Text Generation
70B
Mistral-7B

Sliding window and grouped-query attention for fast, memory-lean inference. Mistral AI, 7B.

Text Generation
7B
Mistral-7B-Instruct-v0.3

Function-calling and instruction following for production and agentic use. Mistral AI, 7B.

Text Generation
7B
Mixtral-8x7B

Sparse MoE: 13B active per token out of 45B total — quality at low per-token cost. Mistral AI, 45B.

Text Generation
45B
Motif Vision 6B Preview

Preview text-to-video from natural language prompts. Motif, 6B.

Text-to-Video
6B
Phi-3-Mini-128K-Instruct

Compact 3.8B instruction model with 128K context tuned for reasoning on constrained hardware. Microsoft, 3.8B.

Text Generation
3.8B
Qwen2.5-32B

General-purpose 32B text model with strong multilingual and code coverage. Alibaba, 32B.

Text Generation
32B
Qwen2.5-7B-Instruct

Tuned for code, math, and structured output at an efficient scale. Alibaba, 7B.

Text Generation
7B
Qwen2.5-Coder-32B

Code-specialist 32B tuned for program synthesis, repair, and structured output. Alibaba, 32B.

Text Generation
32B
Qwen2.5-VL-32B

Vision-language base — documents, charts, and multi-image Q&A. Alibaba, 32B.

Image-Text-to-Text
32B
Qwen2.5-VL-32B-Instruct

Document understanding, chart analysis, and multi-image Q&A. Alibaba, 32B.

Image-Text-to-Text
32B
Qwen2.5-VL-3B-Instruct

Vision-language for low-latency multimodal deployment on constrained hardware. Alibaba, 3B.

Image-Text-to-Text
3B
Qwen2.5-VL-72B

Vision-language base at scale — documents, charts, and scene understanding. Alibaba, 72B.

Image-Text-to-Text
72B
Qwen2.5-VL-72B-Instruct

Vision-language at scale — documents, charts, and scene understanding. Alibaba, 72B.

Image-Text-to-Text
72B
Qwen3-32B

Toggleable chain-of-thought for on-demand deep reasoning. Alibaba, 32B.

Text Generation
32B
Qwen3.6-27B

Mid-scale Qwen3.6 for general reasoning, coding, and multilingual generation. Alibaba, 27B.

Text Generation
27B
Qwen3-Embedding-8B

Higher-capacity multilingual embeddings for retrieval and reranking. Alibaba, 8B.

Embedding
8B
Qwen3-VL-32B-Instruct

Hybrid thinking + vision — documents, charts, UI, and multi-image tasks. Alibaba, 32B.

Image-Text-to-Text
32B
QwQ-32B

Chain-of-thought with self-reflection for math, science, and logic. Alibaba, 32B.

Text Generation
32B
SpeechT5 (TTS task)

Mix-transformer segmentation without positional encoding — accurate at low compute. NVIDIA, 3.8M.

Text-to-Speech
307M
Stable Diffusion 3.5 Large

Unified encoder-decoder for natural text-to-speech synthesis. Microsoft, 307M.

Text-to-Image
8B
SD-XL 1.0-base

MMDiT architecture — strong text rendering and prompt control for image generation. Stability AI, 8B.

Text-to-Image
6.6B
Wan2.2 (I2V)

Image-to-video generation animating a still frame with coherent motion. Alibaba, 14B active.

Image-to-Video
14B
Wan2.2

Causal video transformer for text-to-video with strong motion coherence. Alibaba, 14B.

Text-to-Video
14B
whisper-large-v3

99 languages, 680K hours of training audio — built for robust speech recognition. OpenAI, 1.5B.

Speech-to-Text
1.5B
Z-Image-Turbo

Few-step distilled text-to-image generation. Tongyi-MAI.

Text-to-Image
6B

开个找茬奖金呗

与我们共建开放未来。修复漏洞、增加功能、获取报酬。

查看全部
Sep 16
SFPLOADMACRO produces wrong results under coverage instrumentation (Reciprocal fails 46/153 on Blackhole)
medium
Sep 4
Fix INT_MIN correctness in int32 div, remainder, fmod, and scalar promotion
hard
Sep 4
DRAM-sharded decode matmul: the in0 multicast is one hop per activation shard
hard
Sep 2
binary_ng/ternary: replicate the broadcast row with local NoC copies instead of ~2000 scalar stores per tile
medium
Aug 27
ttnn.prod_bw returns non-finite gradients for zero inputs
medium
Aug 15
ttnn.softplus takes the linear branch at exactly input*beta == threshold (strict < where reference uses <=)
medium
Aug 12
Perf - top-k/sort/moe/sampling index tile is written with 1024 scalar stores per tile on a DM core
hard
Aug 6
Generalize multi_scale_deformable_attn to support D values that are multiples of 16
medium
Aug 5
Fix Blackhole destination-reuse synchronization
hard
Aug 4
ttnn/normalization: sharded softmax accepts a subblock_w larger than Dest and silently returns wrong results
medium

加入我们的开发者社区

获取最新信息，提出问题，查看我们的开源仓库。

GitHubDiscordX