[Doc] Host model recipes in recipes/ (#61302)

Signed-off-by: esmeetu <jasonailu87@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Roy Wang
2026-10-17 09:14:36 +00:00
committed by GitHub
co-authored by Claude Opus 5.5
parent f98ae6a9fd
commit b515648641
193 changed files with 44859 additions and 6 deletions
+109
View File
@@ -0,0 +1,109 @@
meta:
title: "Jais-3-0.9B"
slug: "jais-3-0.9b"
provider: "IFM"
description: "Compact dense model for mathematics, code, instruction following, and STEM tasks"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks:
- text
performance_headline: "1.08B stored parameters including embeddings, with 131K context and native K2-Horizon parsing"
related_recipes:
- "IFM/Jais-3-3.7B"
hardware:
h200: verified
model:
model_id: "IFM/Jais-3-0.9B"
architecture: dense
parameter_count: "1.08B"
active_parameters: "1.08B"
context_length: 131072
base_args:
- "--trust-remote-code"
- "--chat-template-content-format"
- "string"
- "--dtype"
- "bfloat16"
- "--max-model-len"
- "131072"
- "--gpu-memory-utilization"
- "0.85"
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 3
description: "Released BF16 checkpoint; the command starts at an 8K runtime context"
compatible_strategies:
- single_node_tp
- multi_node_tp
- multi_node_tp_pp
hardware_overrides: {}
guide: |
## Overview
Jais-3-0.9B is a compact dense model in the IFM Jais-3
family, with 1.08B stored parameters including embeddings and a 131,072-token
context window.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models such as Jais-3-0.9B
for local experimentation, dense mid-size and large models for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
Jais-3-0.9B's `config.json` already enables YaRN, so the full 131,072-token context needs no RoPE override.
```bash
vllm serve IFM/Jais-3-0.9B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--max-model-len 131072 \
--gpu-memory-utilization 0.85 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
Lower `--max-model-len` (e.g. 8192) to reduce KV-cache memory.
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-0.9B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-0.9B on Hugging Face](https://huggingface.co/IFM/Jais-3-0.9B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
+88
View File
@@ -0,0 +1,88 @@
meta:
title: "Jais-3-3.7B"
slug: "jais-3-3.7b"
provider: "IFM"
description: "Dense model for long-context research, serving, and downstream adaptation"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks: [text]
performance_headline: "5.06B stored parameters including embeddings, with up to 512K context"
related_recipes: ["IFM/Jais-3-0.9B", "IFM/Jais-3-7B"]
hardware:
h200: verified
model:
model_id: "IFM/Jais-3-3.7B"
architecture: dense
parameter_count: "5.06B"
active_parameters: "5.06B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 13
description: "BF16 serving runtime with a 131K context baseline"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
guide: |
## Overview
Jais-3-3.7B is a small dense model in the IFM Jais-3 family,
with 5.06B stored parameters including embeddings and support for up to
524,288 tokens of context.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models for higher-capacity serving.
## Launch Commands
```bash
vllm serve IFM/Jais-3-3.7B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-3.7B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-3.7B on Hugging Face](https://huggingface.co/IFM/Jais-3-3.7B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
+93
View File
@@ -0,0 +1,93 @@
meta:
title: "Jais-3-32B"
slug: "jais-3-32b"
provider: "IFM"
description: "Dense long-context model for reasoning research and high-capacity serving"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks: [text]
performance_headline: "34.78B stored parameters including embeddings, with native vLLM execution and up to 512K context"
related_recipes: ["IFM/Jais-3-7B", "IFM/Jais-3-MoVA-36B-A4B"]
hardware:
h200: verified
model:
model_id: "IFM/Jais-3-32B"
architecture: dense
parameter_count: "34.78B"
active_parameters: "34.78B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 84
description: "BF16 serving runtime with native vLLM execution"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
strategy_overrides:
single_node_tp:
tp: 2
guide: |
## Overview
Jais-3-32B is a large dense model in the IFM Jais-3 family,
with 34.78B stored parameters including embeddings and support for up to
524,288 tokens of context.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models such as Jais-3-32B for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/Jais-3-32B \
--tensor-parallel-size 2 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-32B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-32B on Hugging Face](https://huggingface.co/IFM/Jais-3-32B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
+148
View File
@@ -0,0 +1,148 @@
meta:
title: "Jais-3-375B-A23B"
slug: "jais-3-375b-a23b"
provider: "IFM"
description: "Frontier-scale sparse MoE model for long-context research and high-capacity serving"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks: [text]
performance_headline: "379.17B stored parameters including embeddings / 26.67B active parameters per token"
related_recipes: ["IFM/Jais-3-MoVA-36B-A4B"]
hardware:
h200: verified
mi355x: verified
model:
model_id: "IFM/Jais-3-375B-A23B"
architecture: moe
parameter_count: "379.17B"
active_parameters: "26.67B"
context_length: 524288
# --enable-expert-parallel is not here: in base_args it would reach the AMD
# path too, which deliberately runs pure TP (see strategy_hardware). The TEP
# strategies emit it themselves; multi_node_tp_pp does not, so it carries its
# own copy in strategy_overrides below.
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 911
description: "BF16 serving runtime with an eight-GPU tensor-parallel baseline"
compatible_strategies: [single_node_tep, single_node_tp, multi_node_tep, multi_node_tp_pp]
default_strategy: single_node_tep
# On AMD the expert-parallel dispatch is slower than replicating the experts
# across a pure tensor-parallel shard, so the TEP strategies are opted out and
# single-node TP is the offered path. multi_node_tp_pp is left available: EP
# works there, it is just not the fast default at single-node scale.
strategy_hardware:
single_node_tep:
amd: unsupported
multi_node_tep:
amd: unsupported
hardware_overrides:
amd:
extra_args:
- "--attention-backend"
- "ROCM_AITER_UNIFIED_ATTN"
- "--gpu-memory-utilization"
- "0.9"
extra_env:
VLLM_ROCM_USE_AITER: "1"
NCCL_MIN_NCHANNELS: "112"
strategy_overrides:
single_node_tep:
tp: 8
# multi_node_tp_pp emits no EP flag of its own, and this recipe has always
# run it with expert parallel on. Restored here rather than in base_args so
# it stays off the AMD single-node TP path.
multi_node_tp_pp:
extra_args: ["--enable-expert-parallel"]
guide: |
## Overview
Jais-3-375B-A23B is a frontier-scale mixture-of-expert (MoE) model in the IFM
Jais-3 family, with 379.17B stored parameters including embeddings,
26.67B active parameters per token, and support for up to 524,288 tokens of
context.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models such as Jais-3-375B-A23B for frontier-scale
research and higher-capacity serving.
## Launch Commands
```bash
vllm serve IFM/Jais-3-375B-A23B \
--tensor-parallel-size 8 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--enable-expert-parallel \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
### AMD MI355X
Verified on an 8-GPU MI355X (gfx950) node. AITER supplies the fused MoE and
the unified attention backend:
```bash
export VLLM_ROCM_USE_AITER=1
export NCCL_MIN_NCHANNELS=112
vllm serve IFM/Jais-3-375B-A23B \
--tensor-parallel-size 8 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--max-model-len 131072 \
--gpu-memory-utilization 0.9 \
--attention-backend ROCM_AITER_UNIFIED_ATTN \
--generation-config auto \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-375B-A23B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-375B-A23B on Hugging Face](https://huggingface.co/IFM/Jais-3-375B-A23B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
+89
View File
@@ -0,0 +1,89 @@
meta:
title: "Jais-3-7B"
slug: "jais-3-7b"
provider: "IFM"
description: "Dense model for long-context research, fine-tuning, and cost-conscious deployment"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks: [text]
performance_headline: "9B stored parameters including embeddings, with up to 512K context"
related_recipes: ["IFM/Jais-3-3.7B", "IFM/Jais-3-32B"]
hardware:
h200: verified
model:
model_id: "IFM/Jais-3-7B"
architecture: dense
parameter_count: "9B"
active_parameters: "9B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 22
description: "BF16 serving runtime with a 131K context baseline"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
guide: |
## Overview
Jais-3-7B is a compact dense model in the IFM Jais-3 family,
with 9B stored parameters including embeddings and support for up to 524,288
tokens of context.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models such as Jais-3-7B for
local experimentation, dense mid-size and large models for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/Jais-3-7B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-7B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-7B on Hugging Face](https://huggingface.co/IFM/Jais-3-7B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
@@ -0,0 +1,96 @@
meta:
title: "Jais-3-MoVA-36B-A4B"
slug: "jais-3-mova-36b-a4b"
provider: "IFM"
description: "Sparse MoVA and MoE long-context model for research and production-style serving experiments"
date_added: 2026-10-05
date_updated: 2026-10-05
tasks: [text]
performance_headline: "37.44B stored parameters including embeddings / 5.95B active parameters per token"
related_recipes: ["IFM/Jais-3-32B", "IFM/Jais-3-375B-A23B"]
hardware:
h200: verified
model:
model_id: "IFM/Jais-3-MoVA-36B-A4B"
architecture: moe
parameter_count: "37.44B"
active_parameters: "5.95B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16", "--enable-expert-parallel"]
base_env: {}
features:
reasoning:
description: "Extract Jais-3 reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 90
description: "BF16 serving runtime with native vLLM execution"
compatible_strategies: [single_node_tep, multi_node_tep, multi_node_tp_pp]
default_strategy: single_node_tep
hardware_overrides: {}
strategy_overrides:
single_node_tep:
tp: 2
guide: |
## Overview
Jais-3-MoVA-36B-A4B is a mixture-of-expert (MoE) model in the IFM Jais-3
family. It combines MoVA attention and MoE layers, with 37.44B stored parameters
including embeddings, 5.95B active parameters per token, and support for up
to 524,288 tokens of context.
Jais-3 is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models such as Jais-3-MoVA-36B-A4B for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/Jais-3-MoVA-36B-A4B \
--tensor-parallel-size 2 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--enable-expert-parallel \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/Jais-3-MoVA-36B-A4B",
messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "low"}` — no thinking.
## References
- [IFM/Jais-3-MoVA-36B-A4B on Hugging Face](https://huggingface.co/IFM/Jais-3-MoVA-36B-A4B)
- [Jais 3 Collection](https://huggingface.co/collections/IFM/jais-3)
+110
View File
@@ -0,0 +1,110 @@
meta:
title: "K2-Horizon-0.9B"
slug: "k2-horizon-0.9b"
provider: "IFM"
description: "Compact dense model for mathematics, code, instruction following, and STEM tasks"
date_added: 2026-08-31
date_updated: 2026-08-31
tasks:
- text
performance_headline: "1.08B stored parameters including embeddings, with 131K context and native K2-Horizon parsing"
related_recipes:
- "IFM/K2-Horizon-3.7B"
hardware:
h200: verified
model:
model_id: "IFM/K2-Horizon-0.9B"
architecture: dense
parameter_count: "1.08B"
active_parameters: "1.08B"
context_length: 131072
base_args:
- "--trust-remote-code"
- "--chat-template-content-format"
- "string"
- "--dtype"
- "bfloat16"
- "--max-model-len"
- "131072"
- "--gpu-memory-utilization"
- "0.85"
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 3
description: "Released BF16 checkpoint; the command starts at an 8K runtime context"
compatible_strategies:
- single_node_tp
- multi_node_tp
- multi_node_tp_pp
hardware_overrides: {}
guide: |
## Overview
K2-Horizon-0.9B is a compact dense model in the IFM K2-Horizon
family, with 1.08B stored parameters including embeddings and a 131,072-token
context window.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models such as K2-Horizon-0.9B
for local experimentation, dense mid-size and large models for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
K2-Horizon-0.9B's `config.json` already enables YaRN, so the full 131,072-token context needs no RoPE override.
```bash
vllm serve IFM/K2-Horizon-0.9B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--max-model-len 131072 \
--gpu-memory-utilization 0.85 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
Lower `--max-model-len` (e.g. 8192) to reduce KV-cache memory.
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-0.9B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-0.9B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-0.9B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)
+89
View File
@@ -0,0 +1,89 @@
meta:
title: "K2-Horizon-3.7B"
slug: "k2-horizon-3.7b"
provider: "IFM"
description: "Dense model for long-context research, serving, and downstream adaptation"
date_added: 2026-08-31
date_updated: 2026-09-02
tasks: [text]
performance_headline: "5.06B stored parameters including embeddings, with up to 512K context"
related_recipes: ["IFM/K2-Horizon-0.9B", "IFM/K2-Horizon-7B"]
hardware:
h200: verified
model:
model_id: "IFM/K2-Horizon-3.7B"
architecture: dense
parameter_count: "5.06B"
active_parameters: "5.06B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 13
description: "BF16 serving runtime with a 131K context baseline"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
guide: |
## Overview
K2-Horizon-3.7B is a small dense model in the IFM K2-Horizon family,
with 5.06B stored parameters including embeddings and support for up to
524,288 tokens of context.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models for higher-capacity serving.
## Launch Commands
```bash
vllm serve IFM/K2-Horizon-3.7B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-3.7B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-3.7B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-3.7B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)
+94
View File
@@ -0,0 +1,94 @@
meta:
title: "K2-Horizon-32B"
slug: "k2-horizon-32b"
provider: "IFM"
description: "Dense long-context model for reasoning research and high-capacity serving"
date_added: 2026-08-31
date_updated: 2026-09-02
tasks: [text]
performance_headline: "34.78B stored parameters including embeddings, with native vLLM execution and up to 512K context"
related_recipes: ["IFM/K2-Horizon-7B", "IFM/K2-Horizon-MoVA-36B-A4B"]
hardware:
h200: verified
model:
model_id: "IFM/K2-Horizon-32B"
architecture: dense
parameter_count: "34.78B"
active_parameters: "34.78B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 84
description: "BF16 serving runtime with native vLLM execution"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
strategy_overrides:
single_node_tp:
tp: 2
guide: |
## Overview
K2-Horizon-32B is a large dense model in the IFM K2-Horizon family,
with 34.78B stored parameters including embeddings and support for up to
524,288 tokens of context.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models such as K2-Horizon-32B for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/K2-Horizon-32B \
--tensor-parallel-size 2 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-32B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-32B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-32B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)
@@ -0,0 +1,149 @@
meta:
title: "K2-Horizon-375B-A23B"
slug: "k2-horizon-375b-a23b"
provider: "IFM"
description: "Frontier-scale sparse MoE model for long-context research and high-capacity serving"
date_added: 2026-08-31
date_updated: 2026-09-08
tasks: [text]
performance_headline: "379.17B stored parameters including embeddings / 26.67B active parameters per token"
related_recipes: ["IFM/K2-Horizon-MoVA-36B-A4B"]
hardware:
h200: verified
mi355x: verified
model:
model_id: "IFM/K2-Horizon-375B-A23B"
architecture: moe
parameter_count: "379.17B"
active_parameters: "26.67B"
context_length: 524288
# --enable-expert-parallel is not here: in base_args it would reach the AMD
# path too, which deliberately runs pure TP (see strategy_hardware). The TEP
# strategies emit it themselves; multi_node_tp_pp does not, so it carries its
# own copy in strategy_overrides below.
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 911
description: "BF16 serving runtime with an eight-GPU tensor-parallel baseline"
compatible_strategies: [single_node_tep, single_node_tp, multi_node_tep, multi_node_tp_pp]
default_strategy: single_node_tep
# On AMD the expert-parallel dispatch is slower than replicating the experts
# across a pure tensor-parallel shard, so the TEP strategies are opted out and
# single-node TP is the offered path. multi_node_tp_pp is left available: EP
# works there, it is just not the fast default at single-node scale.
strategy_hardware:
single_node_tep:
amd: unsupported
multi_node_tep:
amd: unsupported
hardware_overrides:
amd:
extra_args:
- "--attention-backend"
- "ROCM_AITER_UNIFIED_ATTN"
- "--gpu-memory-utilization"
- "0.9"
extra_env:
VLLM_ROCM_USE_AITER: "1"
NCCL_MIN_NCHANNELS: "112"
strategy_overrides:
single_node_tep:
tp: 8
# multi_node_tp_pp emits no EP flag of its own, and this recipe has always
# run it with expert parallel on. Restored here rather than in base_args so
# it stays off the AMD single-node TP path.
multi_node_tp_pp:
extra_args: ["--enable-expert-parallel"]
guide: |
## Overview
K2-Horizon-375B-A23B is a frontier-scale mixture-of-expert (MoE) model in the IFM
K2-Horizon family, with 379.17B stored parameters including embeddings,
26.67B active parameters per token, and support for up to 524,288 tokens of
context.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models such as K2-Horizon-375B-A23B for frontier-scale
research and higher-capacity serving.
## Launch Commands
```bash
vllm serve IFM/K2-Horizon-375B-A23B \
--tensor-parallel-size 8 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--enable-expert-parallel \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
### AMD MI355X
Verified on an 8-GPU MI355X (gfx950) node. AITER supplies the fused MoE and
the unified attention backend:
```bash
export VLLM_ROCM_USE_AITER=1
export NCCL_MIN_NCHANNELS=112
vllm serve IFM/K2-Horizon-375B-A23B \
--tensor-parallel-size 8 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--max-model-len 131072 \
--gpu-memory-utilization 0.9 \
--attention-backend ROCM_AITER_UNIFIED_ATTN \
--generation-config auto \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-375B-A23B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-375B-A23B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-375B-A23B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)
+90
View File
@@ -0,0 +1,90 @@
meta:
title: "K2-Horizon-7B"
slug: "k2-horizon-7b"
provider: "IFM"
description: "Dense model for long-context research, fine-tuning, and cost-conscious deployment"
date_added: 2026-08-31
date_updated: 2026-09-02
tasks: [text]
performance_headline: "9B stored parameters including embeddings, with up to 512K context"
related_recipes: ["IFM/K2-Horizon-3.7B", "IFM/K2-Horizon-32B"]
hardware:
h200: verified
model:
model_id: "IFM/K2-Horizon-7B"
architecture: dense
parameter_count: "9B"
active_parameters: "9B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16"]
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 22
description: "BF16 serving runtime with a 131K context baseline"
compatible_strategies: [single_node_tp, multi_node_tp, multi_node_tp_pp]
hardware_overrides: {}
guide: |
## Overview
K2-Horizon-7B is a compact dense model in the IFM K2-Horizon family,
with 9B stored parameters including embeddings and support for up to 524,288
tokens of context.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models such as K2-Horizon-7B for
local experimentation, dense mid-size and large models for high-quality
research workloads, and mixture-of-expert (MoE) models for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/K2-Horizon-7B \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-7B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-7B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-7B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)
@@ -0,0 +1,97 @@
meta:
title: "K2-Horizon-MoVA-36B-A4B"
slug: "k2-horizon-mova-36b-a4b"
provider: "IFM"
description: "Sparse MoVA and MoE long-context model for research and production-style serving experiments"
date_added: 2026-08-31
date_updated: 2026-09-03
tasks: [text]
performance_headline: "37.44B stored parameters including embeddings / 5.95B active parameters per token"
related_recipes: ["IFM/K2-Horizon-32B", "IFM/K2-Horizon-375B-A23B"]
hardware:
h200: verified
model:
model_id: "IFM/K2-Horizon-MoVA-36B-A4B"
architecture: moe
parameter_count: "37.44B"
active_parameters: "5.95B"
context_length: 524288
base_args: ["--trust-remote-code", "--chat-template-content-format", "string", "--dtype", "bfloat16", "--enable-expert-parallel"]
base_env: {}
features:
reasoning:
description: "Extract K2-Horizon reasoning into message.reasoning"
args: ["--reasoning-parser", "k2_horizon"]
tool_calling:
description: "Enable automatic tool choice with the K2-Horizon parser"
args: ["--enable-auto-tool-choice", "--tool-call-parser", "k2_horizon"]
variants:
default:
precision: bf16
vram_minimum_gb: 90
description: "BF16 serving runtime with native vLLM execution"
compatible_strategies: [single_node_tep, multi_node_tep, multi_node_tp_pp]
default_strategy: single_node_tep
hardware_overrides: {}
strategy_overrides:
single_node_tep:
tp: 2
guide: |
## Overview
K2-Horizon-MoVA-36B-A4B is a mixture-of-expert (MoE) model in the IFM K2-Horizon
family. It combines MoVA attention and MoE layers, with 37.44B stored parameters
including embeddings, 5.95B active parameters per token, and support for up
to 524,288 tokens of context.
K2-Horizon is an open-weight IFM model family built for transparent
foundation-model research, staged checkpoint analysis, and practical
deployment. The series spans compact dense models for local experimentation,
dense mid-size and large models for high-quality research workloads, and
mixture-of-expert (MoE) models such as K2-Horizon-MoVA-36B-A4B for higher-capacity
serving.
## Launch Commands
```bash
vllm serve IFM/K2-Horizon-MoVA-36B-A4B \
--tensor-parallel-size 2 \
--trust-remote-code \
--chat-template-content-format string \
--dtype bfloat16 \
--enable-expert-parallel \
--reasoning-parser k2_horizon \
--enable-auto-tool-choice \
--tool-call-parser k2_horizon
```
## Client usage
```python
from openai import OpenAI
client = OpenAI(
api_key="EMPTY", base_url="http://localhost:8000/v1", timeout=3600
)
resp = client.chat.completions.create(
model="IFM/K2-Horizon-MoVA-36B-A4B",
messages=[{"role": "user", "content": "Give me three primes above 100."}],
temperature=1.0, top_p=0.95, max_tokens=2048,
)
print(resp.choices[0].message.content)
```
### Thinking modes
The model supports selectable thinking through `chat_template_kwargs`, per
request or server-wide via `--default-chat-template-kwargs`:
- `{"reasoning_effort": "high"}` — full thinking (default).
- `{"reasoning_effort": "medium"}` — faster thinking.
- `{"reasoning_effort": "low"}` — fastest thinking.
## References
- [IFM/K2-Horizon-MoVA-36B-A4B on Hugging Face](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B)
- [K2 Horizon Collection](https://huggingface.co/collections/IFM/k2-horizon)