Model reference · open weights
K2 is an open-weight language model from IFM. K2 (FP16) weighs 131 GB; the smallest configuration that runs it is 4× L40S 48 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the IFM/K2 model card, 2026-10-01
What it is
| Released by | IFM |
|---|---|
| Released | 2024-04-17 |
| Parameters | 65.3B |
| VRAM | 131 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory |
|---|---|---|---|
| 8K each | |||
| RTX 3060 12 GB … DGX Spark (GB10) 128 GB unified 10 smaller cards | |||
| H200 141 GB | 1K | 138 GB | |
| B200 180 GB | 1 | all 8K | 176 GB |
| 4× L40S 48 GB tensor parallel | 1 | all 8K | 44.0 GB a card |
| 4× A100 80 GB tensor parallel | 8 | all 8K | 78.2 GB a card |
| 2× B200 180 GB tensor parallel | 9 | all 8K | 176 GB a card |
| 4× RTX PRO 6000 Blackwell 96 GB tensor parallel | 10 | all 8K | 93.8 GB a card |
| 4× H200 141 GB tensor parallel | 18 | all 8K | 138 GB a card |
| 8× H100 80 GB tensor parallel | 21 | all 8K | 78.1 GB a card |
| Requests at once | 8K tokens each |
|---|---|
| 1 | 153 GB |
| 5 | 239 GB |
| 8 | 303 GB |
| 16 | 475 GB |
| 32 | 819 GB |
| 64 | 1506 GB |
One card, with vLLM's small-card settings.
From the model card
LLM360 demystifies the training recipe used for Llama 2 70B with K2. K2 is fully transparent, meaning we’ve open-sourced all artifacts, including code, data, model checkpoints, intermediate results, and more.
K2 was developed as a collaboration between MBZUAI, Petuum, and LLM360.
The LLM360 Performance and Evaluation Collection is a robust evaluations set consisting of general and domain specific evaluations to assess model knowledge and function.
Evaluations include standard best practice benchmarks, medical, math, and coding knowledge. More about the evaluations can be found here.
Detailed analysis can be found on the K2 Weights and Biases project here
| Evaluation | Score | Raw Score |
|---|---|---|
| IFEval | 22.52 | 23 |
| BBH | 28.22 | 50 |
| Math Lvl 5 | 2.04 | 2 |
| GPQA | 3.58 | 28 |
| MUSR | 8.55 | 40 |
| MMLU-PRO | 22.27 | 30 |
| Average | 14.53 | 35.17 |
The K2 gallery allows one to browse the output of various prompts on intermediate K2 checkpoints, which provides an intuitive understanding on how the model develops and improves over time. This is inspired by The Bloom Book.
The following data mix was used to train K2 and achieve results in line with Llama 2 70B.
The full data sequence can be found here
| Dataset | Starting Tokens | Multiplier | Total Tokens | % of Total |
|---|---|---|---|---|
| dm-math | 4.33B | 3x | 13B | 1% |
| pubmed-abstracts | 4.77B | 3x | 14.3B | 1.1% |
| uspto | 4.77B | 3x | 14.3B | 1.1% |
| pubmed-central | 26B | 1x | 26B | 2% |
| redpajama.arxiv | 27.3B | 1x | 27.3B | 2.1% |
| starcoder.spm | 67.6B | 0.5x | 33.8B | 2.6% |
| starcoder.fim | 67.6B | 0.5x | 33.8B | 2.6% |
| redpajama.stackexchange | 61.1B | 1x | 61.1B | 4.7% |
| starcoder | 132.6B | 0.5x | 66.3B | 5.1% |
| pile-of-law | 76.7B | 1x | 76.7B | 5.9% |
| redpajama.book | 80.6B | 1x | 80.6B | 6.2% |
| s2orc | 107.9B | 1x | 107.9B | 8.3% |
| redpajama.wikipedia | 22.1B | 6x | 132.6B | 10.2% |
| refinedweb | 612.3B | 1x | 612.3B | 47.1% |
| Totals | - | - | 1.3T | 100% |
| Checkpoints | |
|---|---|
| Checkpoint 380 | Checkpoint 375 |
| Checkpoint 379 | Checkpoint 374 |
| Checkpoint 378 | Checkpoint 373 |
| Checkpoint 377 | Checkpoint 372 |
| Checkpoint 376 | Checkpoint 371 |
| Checkpoints | |
|---|---|
| Checkpoint 360 | Checkpoint 355 |
| Checkpoint 359 | Checkpoint 354 |
| Checkpoint 358 | Checkpoint 353 |
| Checkpoint 357 | Checkpoint 352 |
| Checkpoint 356 | Checkpoint 351 |
[to find all branches: git branch -a]
We provide step-by-step reproducation tutorials for tech enthusiasts, AI practitioners and academic or industry researchers who want to learn pretraining techniques here.
We provide step-by-step finetuning tutorials for tech enthusiasts, AI practitioners and academic or industry researchers [here](https://www.llm360.ai/develo
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.