K2 Horizon 7B
K2 Horizon 7B is a text-generation model from IFM/K2-Horizon-7B, licensed under apache-2.0, with safetensors and text-generation tags. According to the official model card, it is a medium dense member of the K2-Horizon family: a 7B-core decoder-only model with a 512K context window. The card states a native 524,288-token context from midtraining onward, describes diffusion adapters for faster inference, notes released intermediate checkpoints, and says training data, recipe, training code, and evaluation resources are public. It lists datasets IFM/K2-Horizon-Pretrain-Data and IFM/K2-Horizon-Midtrain-Data, language English, and a transformers library name.
At a glance
- Parameters
- 9.00B
- Architecture
- k2_horizon
- Context length
- 524,288
- License
- apache-2.0
- Software
- SGLang, vLLM, Transformers
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Text generation with a decoder-only model
- Long-context tasks up to a native 524,288-token context window
- Studying capability changes across released intermediate checkpoints
- Evaluating the model across agentic, coding, long-context, and reasoning benchmarks
- Exploring faster inference via diffusion adapters
Tradeoffs
- Only text-generation and English language are stated; no other tasks or languages are specified.
- No explicit hardware requirements, memory needs, or performance guarantees are provided in the source data.
- No compatibility details are given beyond the transformers library name.
- The model is described as medium dense and 7B-core; no quantization or deployment details are supplied in the provided facts.
- Benchmark table in the source is incomplete in the provided excerpt, so full comparative results and reference model details cannot be summarized.
Ways to run it
SGLang · Linux
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
sglang serve --model-path IFM/K2-Horizon-7B --revision 30d38fecf8a609873ae73a617f5c714286e1f565 --tp 1 --dtype bfloat16 --attention-backend fa3 --reasoning-parser k2_horizon --host 0.0.0.0 --port 30000vLLM · Linux
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
vllm serve IFM/K2-Horizon-7B --trust-remote-code --dtype bfloat16 --tensor-parallel-size 1 --reasoning-parser k2_horizon --enable-auto-tool-choice --tool-call-parser k2_horizonTransformers · See official guide
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "IFM/K2-Horizon-7B"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_id, device_map="auto", dtype="bfloat16", low_cpu_mem_usage=True, trust_remote_code=True
)
inputs = tokenizer("Explain why long-context evaluation is difficult.", return_tensors="pt").to(model.device)
inputs.pop("token_type_ids", None)
outputs = model.generate(**inputs, max_new_tokens=32768, temperature=1.0, top_p=0.95, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))Explore its uses
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
A multimodal reasoning model with a compressed key-value cache, published as open weights.
Qwen3.8-Flash-Next
An experimental open-weight multimodal model with sparse attention and 262K native context.