Model reference · open weights
NCP_ArchPreview_dolma3_8.9B_Stage2 is an open-weight language model from ArchSpace-Collection. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.
Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.
What it is
| Released by | ArchSpace-Collection |
|---|---|
| Type | Language models |
| Task | Text gen |
| Parameters (lead) | 8.9B |
| Context | 8k tokens |
| Runs with | transformers |
| Based on | ArchSpace-Collection/NCP_ArchPreview_dolma3_8.9B_Stage1 |
| Released | 2026-08-27 |
| Popularity | 920 downloads / month |
| Licence | Open weights |
About
Model collection | Technical report (arXiv) | HF Papers | Training code (coming soon) | Evaluation code
NCP-ArchPreview Stage 2 continues the Stage 1 model on Dolma 3 Dolmino, following the second stage of the OLMo-3 data curriculum. It retains joint Next Token Prediction (NTP) and Next Concept Prediction (NCP), allowing the model to refine token-level capabilities and its learned concept space during continued pretraining.
Developed by The NCP Team at Shanghai AI Lab and LUMIA Lab, Shanghai Jiao Tong University, the model has approximately 8.94B parameters and an 8,192-token training context. Stage 2 remains a base-model release for completion, evaluation, and further adaptation.
The report also finds that Stage 2 reaches the OLMo-3-7B final training loss with 66.2% of the training tokens, corresponding to 1.51x convergence in token budget. This is a training-loss comparison, not an inference throughput measurement.
| Variant | Hugging Face repository |
|---|---|
| V1 | NCP_ArchPreview_dolma3_8.9B_Stage2_v1 |
| V2 | NCP_ArchPreview_dolma3_8.9B_Stage2_v2 |
| V3 | NCP_ArchPreview_dolma3_8.9B_Stage2_v3 |
These labels identify alternative Stage 2 data recipes. Higher version numbers do not imply a stronger model: although aggregate training loss decreases from V1 to V3, several downstream capabilities decline. The quick-start example uses V1, which is the strongest of these recipes across the free-form capabilities evaluated in Appendix C.
The following values are the report's main Stage 2 comparison with OLMo-3-7B and should not be assigned to all three released variants. Scores are percentages; deltas are absolute percentage points.
| Metric | OLMo-3-7B Stage 2 | NCP-ArchPreview Stage 2 | Delta |
|---|---|---|---|
| Overall AVG | 56.98 | 57.57 | +0.59 |
| MMLU | 66.66 | 68.48 | +1.82 |
| GSM8K | 79.68 | 83.02 | +3.34 |
| GSM-Symbolic | 57.32 | 60.32 | +3.00 |
| MATH-500 | 43.44 | 43.91 | +0.47 |
| HumanEval | 49.31 | 45.62 | -3.69 |
| MBPP | 48.98 | 50.85 | +1.87 |
| ARC-Challenge | 85.49 | 83.28 | -2.21 |
| PIQA | 78.35 | 81.45 | +3.10 |
| Domain average | OLMo-3-7B Stage 2 | NCP-ArchPreview Stage 2 |
|---|---|---|
| MMLU family | 58.32 | 59.94 |
| Mathematics | 55.63 | 57.39 |
| Code | 39.42 | 38.77 |
| Multiple-choice STEM | 89.65 | 88.67 |
| Multiple-choice non-STEM | 76.85 | 77.76 |
| GenQA | 53.49 | 54.04 |
Likelihood is reported separately in bits per UTF-8 byte (BPB), where lower is better.
| Likelihood metric | OLMo-3-7B Stage 2 | NCP-ArchPreview Stage 2 |
|---|---|---|
| BPB AVG | 0.793 | 0.763 |
The main Stage 2 model improves the overall average while showing a mixed profile across individual tasks, including lower HumanEval and ARC-Challenge scores. The report discusses a possible mismatch between the continued-training mixture and downstream domains; lower aggregate loss alone does not establish better performance on every task.
All three Stage 2 variants share the same inference and sampling configuration within this comparison.
| Benchmark | v1 | v2 | v3 |
|---|---|---|---|
| HumanEval | 45.60 | 42.19 | 39.96 |
| MBPP | 50.91 | 49.34 | 46.60 |
| MATH-500 | 43.74 | 41.66 | 37.21 |
| Minerva | 42.20 | 40.39 | 36.70 |
| MMLU-STEM | 61.84 | 59.85 | 57.72 |
| BBH | 63.23 | 62.90 | 60.44 |
| HellaSwag | 66.40 | 67.30 | 67.25 |
For recipe selection, the report treats V2 and V3 as effectively tied on HellaSwag because they differ by only 0.05 percentage points. Its held-out proxy analysis supports the V1 > V2 > V3 ordering on four free-form capability measurements. This evidence compares complete recipes and does not isolate the contribution of an individual data source.
The main comparison follows the OLMo evaluation protocol in OLMo-Core. Appendix E and Table 13 specify the per-task settings, including:
The sample count is the number of generated completions per problem, not the inference batch size. Overall AVG is the unweighted mean of 26 constituent benchmark scores, excluding the aggregate MMLU row and domain averages. The ten-benchmark BPB AVG is calculated separately and is excluded from Overall AVG. See the evaluation configuration for the corresponding release assets
From the published model card. Full card on the HuggingFace links in the sidebar.
Using it via the API
Once AxForge deploys ncp-archpreview-dolma3-8-9b-stage2 for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (ncp-archpreview-dolma3-8-9b-stage2 below is illustrative; you get the exact model name on deployment.)
$ curl -sS https://api.axforge.ai/v1/chat/completions \
-H "Authorization: Bearer $AXFORGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"ncp-archpreview-dolma3-8-9b-stage2","messages":[{"role":"user","content":"Hello"}]}'
Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.