{ "@context": "https://schema.org", "@id": "https://aggr.aymericbeaumet.com/items/zml-ai/2026-10-29-zml-llmd-20261029-0/#webpage", "@type": [ "WebPage", "ArchiveComponent" ], "articleBody": "[Back to posts](https://zml.ai/posts/)\n\nArticle\n\nLe Chonk.\n\nOctober 29, 2026 inference / serving / mistral / chonk / glm / lfm\n\nWe are pleased to announce the release of `zml/llmd 20261029.0`. This week, we are bringing more models:\n\n- [Mistral Large 4](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B)\n- [GLM 5.2](https://huggingface.co/collections/zai-org/glm-52) and [GLM 5.3](https://huggingface.co/collections/zai-org/glm-53) series\n- [LFM 3](https://huggingface.co/LiquidAI/LFM3-1.7B) series\n\n## Mistral Large 4 – Le Chonk ![](https://zml.ai/posts/llmd-20261029.0/big-cat-icon.svg)\n\nIn coordination with Mistral, we are pleased to announce Day 0 support for [Mistral Large 4](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B).\n\n> Mistral Large 4 Le Chonk (Le Chonk) is a 1-trillion-parameter multimodal model, now available with improved coding, cybersecurity, and reasoning efficiency. It leads open-weight benchmarks in coding and defensive security while offering three reasoning modes for cost-performance balance. Fully open-weight and European-trained, it empowers sovereign AI with customizable, self-hosted deployment options.\n\n[See the press release](https://mistral.ai/news/mistral-large-4-oss/).\n\nThe native DSpark Speculative decoding is also supported via the `--speculative-drafter-model=native` flag.\n\nDecode throughput, mistralai/Mistral-Large-4-1T-A52B, bs=1, 4xGB300 (tok/s)\n\n```\n$ docker run -p 8000:8000 --shm-size=32GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n --model=hf://mistralai/Mistral-Large-4-1T-A52B\n```\n\nSadly, we could not bring it to AMD due to [a bug in the `amdgpu` driver that’s now been fixed but not released](https://www.mail-archive.com/amd-gfx@lists.freedesktop.org/msg152088.html). We are actively [monitoring the situation](https://zml.ai/posts/llmd-20261029.0/monitoring_the_situation.jpeg).\n\nEven though it should work, we haven’t tested it on Google TPUs because, again, we could not get any.\n\n## GLM 5.2 and GLM 5.3\n\nThis release also brings support for the [GLM 5.2](https://huggingface.co/collections/zai-org/glm-52) and [GLM 5.3](https://huggingface.co/collections/zai-org/glm-53) series of models.\n\n## LFM 3\n\nThis release brings support for the [LFM 3](https://huggingface.co/LiquidAI/LFM3-1.7B) series of models, in all flavours (BF16, FP8 and NVFP4).\n\nThis model is great at agentic tasks and following instructions. Combined with our NVFP4 support on Metal, it makes for a great fully local combination, especially when paired with the [Lightpanda MCP](https://lightpanda.io/docs/usage/mcp):\n\nDecode throughput, LiquidAI/LFM3-1.7B, bs=1, zml/llmd 20261029.0, M3 Max (tok/s)\n\n![](https://zml.ai/posts/llmd-20261029.0/dsh.png)\n\n## Try it today\n\n## NVIDIA\n\n```bash\ndocker run -p 8000:8000 --shm-size=256GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n --model=hf://LiquidAI/LFM3-1.7B-FP8\n```\n\n## AMD\n\n```bash\ndocker run -p 8000:8000 --shm-size=256GB --device=/dev/kfd --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:rocm \\\n --model=hf://LiquidAI/LFM3-1.7B-FP8\n```\n\n## Intel\n\n```bash\ndocker run -p 8000:8000 --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:oneapi \\\n --model=hf://LiquidAI/LFM3-1.7B\n```\n\n## Google TPU\n\n```bash\ndocker run --net=host --privileged -e HF_TOKEN -it zmlai/llmd:tpu \\\n --model=hf://LiquidAI/LFM3-1.7B\n```\n\n```bash\nbrew install zml/zml/llmd\nllmd --model=hf://LiquidAI/LFM3-1.7B-FP8 --batch-size=4\n```\n", "authors": [], "captured_at": "2026-10-29T20:30:27.834895129+00:00", "category": "ai", "content_html": "

Back to posts

\n

Article

\n

Le Chonk.

\n

October 29, 2026 inference / serving / mistral / chonk / glm / lfm

\n

We are pleased to announce the release of zml/llmd 20261029.0. This week, we are bringing more models:

\n\n

Mistral Large 4 – Le Chonk \"\"

\n

In coordination with Mistral, we are pleased to announce Day 0 support for Mistral Large 4.

\n
\n

Mistral Large 4 Le Chonk (Le Chonk) is a 1-trillion-parameter multimodal model, now available with improved coding, cybersecurity, and reasoning efficiency. It leads open-weight benchmarks in coding and defensive security while offering three reasoning modes for cost-performance balance. Fully open-weight and European-trained, it empowers sovereign AI with customizable, self-hosted deployment options.

\n
\n

See the press release.

\n

The native DSpark Speculative decoding is also supported via the --speculative-drafter-model=native flag.

\n

Decode throughput, mistralai/Mistral-Large-4-1T-A52B, bs=1, 4xGB300 (tok/s)

\n
$ docker run -p 8000:8000 --shm-size=32GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n    --model=hf://mistralai/Mistral-Large-4-1T-A52B\n
\n

Sadly, we could not bring it to AMD due to a bug in the amdgpu driver that’s now been fixed but not released. We are actively monitoring the situation.

\n

Even though it should work, we haven’t tested it on Google TPUs because, again, we could not get any.

\n

GLM 5.2 and GLM 5.3

\n

This release also brings support for the GLM 5.2 and GLM 5.3 series of models.

\n

LFM 3

\n

This release brings support for the LFM 3 series of models, in all flavours (BF16, FP8 and NVFP4).

\n

This model is great at agentic tasks and following instructions. Combined with our NVFP4 support on Metal, it makes for a great fully local combination, especially when paired with the Lightpanda MCP:

\n

Decode throughput, LiquidAI/LFM3-1.7B, bs=1, zml/llmd 20261029.0, M3 Max (tok/s)

\n

\"\"

\n

Try it today

\n

NVIDIA

\n
docker run -p 8000:8000 --shm-size=256GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n    --model=hf://LiquidAI/LFM3-1.7B-FP8\n
\n

AMD

\n
docker run -p 8000:8000 --shm-size=256GB --device=/dev/kfd --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:rocm \\\n    --model=hf://LiquidAI/LFM3-1.7B-FP8\n
\n

Intel

\n
docker run -p 8000:8000 --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:oneapi \\\n    --model=hf://LiquidAI/LFM3-1.7B\n
\n

Google TPU

\n
docker run --net=host --privileged -e HF_TOKEN -it zmlai/llmd:tpu \\\n    --model=hf://LiquidAI/LFM3-1.7B\n
\n
brew install zml/zml/llmd\nllmd --model=hf://LiquidAI/LFM3-1.7B-FP8 --batch-size=4\n
\n", "content_markdown": "[Back to posts](https://zml.ai/posts/)\n\nArticle\n\nLe Chonk.\n\nOctober 29, 2026 inference / serving / mistral / chonk / glm / lfm\n\nWe are pleased to announce the release of `zml/llmd 20261029.0`. This week, we are bringing more models:\n\n- [Mistral Large 4](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B)\n- [GLM 5.2](https://huggingface.co/collections/zai-org/glm-52) and [GLM 5.3](https://huggingface.co/collections/zai-org/glm-53) series\n- [LFM 3](https://huggingface.co/LiquidAI/LFM3-1.7B) series\n\n## Mistral Large 4 – Le Chonk ![](https://zml.ai/posts/llmd-20261029.0/big-cat-icon.svg)\n\nIn coordination with Mistral, we are pleased to announce Day 0 support for [Mistral Large 4](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B).\n\n> Mistral Large 4 Le Chonk (Le Chonk) is a 1-trillion-parameter multimodal model, now available with improved coding, cybersecurity, and reasoning efficiency. It leads open-weight benchmarks in coding and defensive security while offering three reasoning modes for cost-performance balance. Fully open-weight and European-trained, it empowers sovereign AI with customizable, self-hosted deployment options.\n\n[See the press release](https://mistral.ai/news/mistral-large-4-oss/).\n\nThe native DSpark Speculative decoding is also supported via the `--speculative-drafter-model=native` flag.\n\nDecode throughput, mistralai/Mistral-Large-4-1T-A52B, bs=1, 4xGB300 (tok/s)\n\n```\n$ docker run -p 8000:8000 --shm-size=32GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n --model=hf://mistralai/Mistral-Large-4-1T-A52B\n```\n\nSadly, we could not bring it to AMD due to [a bug in the `amdgpu` driver that’s now been fixed but not released](https://www.mail-archive.com/amd-gfx@lists.freedesktop.org/msg152088.html). We are actively [monitoring the situation](https://zml.ai/posts/llmd-20261029.0/monitoring_the_situation.jpeg).\n\nEven though it should work, we haven’t tested it on Google TPUs because, again, we could not get any.\n\n## GLM 5.2 and GLM 5.3\n\nThis release also brings support for the [GLM 5.2](https://huggingface.co/collections/zai-org/glm-52) and [GLM 5.3](https://huggingface.co/collections/zai-org/glm-53) series of models.\n\n## LFM 3\n\nThis release brings support for the [LFM 3](https://huggingface.co/LiquidAI/LFM3-1.7B) series of models, in all flavours (BF16, FP8 and NVFP4).\n\nThis model is great at agentic tasks and following instructions. Combined with our NVFP4 support on Metal, it makes for a great fully local combination, especially when paired with the [Lightpanda MCP](https://lightpanda.io/docs/usage/mcp):\n\nDecode throughput, LiquidAI/LFM3-1.7B, bs=1, zml/llmd 20261029.0, M3 Max (tok/s)\n\n![](https://zml.ai/posts/llmd-20261029.0/dsh.png)\n\n## Try it today\n\n## NVIDIA\n\n```bash\ndocker run -p 8000:8000 --shm-size=256GB --gpus=all -e HF_TOKEN -it zmlai/llmd:cuda \\\n --model=hf://LiquidAI/LFM3-1.7B-FP8\n```\n\n## AMD\n\n```bash\ndocker run -p 8000:8000 --shm-size=256GB --device=/dev/kfd --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:rocm \\\n --model=hf://LiquidAI/LFM3-1.7B-FP8\n```\n\n## Intel\n\n```bash\ndocker run -p 8000:8000 --device=/dev/dri -e HF_TOKEN -it zmlai/llmd:oneapi \\\n --model=hf://LiquidAI/LFM3-1.7B\n```\n\n## Google TPU\n\n```bash\ndocker run --net=host --privileged -e HF_TOKEN -it zmlai/llmd:tpu \\\n --model=hf://LiquidAI/LFM3-1.7B\n```\n\n```bash\nbrew install zml/zml/llmd\nllmd --model=hf://LiquidAI/LFM3-1.7B-FP8 --batch-size=4\n```\n", "dateCreated": "2026-10-29T20:30:27.834895129+00:00", "date_created": "2026-10-29T07:00:00+00:00", "date_modified": "2026-10-29T07:00:00+00:00", "date_published": "2026-10-29T07:00:00+00:00", "description": "We are pleased to announce the release of zml/llmd 20261029.0. This week, we are bringing more models: Mistral Large 4 GLM 5.2 and GLM 5.3 series LFM 3 series Mistral Large 4 – Le Chonk In coordination with Mistral, we are pleased to…", "external_url": "https://zml.ai/posts/llmd-20261029.0/", "git": { "commit": "20068632b65d2830a2d54f25ca1d90d697bea2c0", "path": "items/zml-ai/2026/10/2026-10-29-zml-llmd-20261029-0.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/20068632b65d2830a2d54f25ca1d90d697bea2c0/items/zml-ai/2026/10/2026-10-29-zml-llmd-20261029-0.md" }, "id": "urn:aggr:item:fdff083557df1a5df51e11b711dd25f67673ddfb", "isBasedOn": { "@id": "https://zml.ai/posts/llmd-20261029.0/" }, "mainEntity": { "@id": "https://zml.ai/posts/llmd-20261029.0/", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/zml-ai/2026-10-29-zml-llmd-20261029-0/#webpage" }, "dateModified": "2026-10-29T07:00:00+00:00", "datePublished": "2026-10-29T07:00:00+00:00", "headline": "zml/llmd 20261029.0", "image": { "@type": "ImageObject", "caption": "zml/llmd 20261029.0 — Le Chonk.", "height": 90, "url": "https://zml.ai/og/background_hu_22c0fbe8ede75b88.png", "width": 160 }, "timeRequired": "PT2M", "url": "https://zml.ai/posts/llmd-20261029.0/", "wordCount": 415 }, "name": "Archived snapshot: zml/llmd 20261029.0", "preview": { "alt": "zml/llmd 20261029.0 — Le Chonk.", "color": null, "height": 90, "placeholder": { "data_url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=", "hash": "" }, "url": "https://zml.ai/og/background_hu_22c0fbe8ede75b88.png", "width": 160 }, "reading_minutes": 2, "replicated_at": null, "source": "zml-ai", "source_name": "ZML - Model to Metal", "summary": "We are pleased to announce the release of zml/llmd 20261029.0. This week, we are bringing more models: Mistral Large 4 GLM 5.2 and GLM 5.3 series LFM 3 series Mistral Large 4 – Le Chonk In coordination with Mistral, we are pleased to announce Day 0 support for Mistral Large 4. Mistral Large 4 Le Chonk (Le Chonk) is a 1-trillion-parameter multimodal model, now available with improved coding, cybersecurity, and reasoning efficiency. It leads open-weight benchmarks in coding and defensive security while offering three reasoning modes for cost-performance balance. Fully open-weight and European-trained, it empowers sovereign AI with customizable, self-hosted deployment options.", "tags": [], "temporalCoverage": "2026-10-29T20:30:27.834895129+00:00", "title": "zml/llmd 20261029.0", "url": "https://aggr.aymericbeaumet.com/items/zml-ai/2026-10-29-zml-llmd-20261029-0/", "word_count": 415 }