---
license: cc-by-4.0
task_categories:
- question-answering
- text-generation
- text2text-generation
tags:
- sft
- instruction-tuning
- agentic
- reverse-engineering
- cad
- svg
- long-context
- synthetic
source_datasets: sft-forge
pretty_name: sft-forge ensemble v1.8
---

# sft-forge-ensemble v1.8

One pareto-pruned, token-rebalanced SFT mixture built by **sft-forge**.
381,768 rows · **2,958,594,172 real gpt2 tokens** · fine-grained tags on every row.

## Token mix by domain (real gpt2, kept rows)
| domain | gpt2 tokens | share |
|---|---|---|
| reasoning_math | 1,308,839,151 | 40.6% |
| reasoning_sci | 441,392,642 | 13.7% |
| agentic_swe | 325,536,103 | 10.1% |
| code_agent | 245,578,724 | 7.6% |
| agentic_tool | 219,243,982 | 6.8% |
| general_sft | 191,070,683 | 5.9% |
| reverse_engineering | 151,780,077 | 4.7% |
| vision_svg | 64,938,936 | 2.0% |
| cad_3d | 9,380,817 | 0.3% |
| style_house | 325,961 | 0.0% |
| vision_image | 294,309 | 0.0% |
| vision_plot | 212,787 | 0.0% |

## Sources
| source | rows | gpt2 tokens |
|---|---|---|
| openmath_cot | 66787 | 535,300,110 |
| opensci | 65779 | 389,184,121 |
| openmath_tir | 165000 | 1,439,717,885 |
| openr1_math | 50000 | 294,774,160 |
| nemotron_sci | 65000 | 96,867,435 |
| ultrachat | 38000 | 45,835,580 |
| svg_code | 8000 | 4,131,989 |
| mcp_atlas | 462 | 7,583,866 |
| pyromind_long | 1000 | 58,991,782 |
| pyromind_short | 1932 | 14,639,677 |
| toolmind | 13351 | 171,755,540 |
| ultradata_tooluse | 2095 | 13,520,358 |
| nemotron_swe | 4702 | 271,084,502 |
| swesmith | 5174 | 164,528,053 |
| coderforge | 2184 | 199,432,411 |
| nemotron_cuda | 2276 | 23,010,820 |
| ultradata_code | 4200 | 276,443,679 |
| agenticsft | 500 | 8,115,596 |
| demyagent | 500 | 1,179,050 |
| ultradata_general | 2692 | 69,243,262 |
| ultradata_search | 2576 | 76,121,807 |
| ollvm_re | 704 | 115,095,930 |
| cadforge_3d | 14 | 35,785 |
| cadforge_cq | 3 | 9,347 |
| multiturn | 192 | 194,013 |
| unified | 758 | 148,433 |
| vision_docmatix | 1617 | 225,903 |
| vision_forge | 1232 | 68,406 |
| plot2code | 379 | 212,787 |
| sophia_svg | 23921 | 57,242,033 |
| icons_forge | 9949 | 3,564,914 |
| re_synth_forge | 6059 | 17,042,338 |
| cad_synth_forge | 5912 | 1,313,781 |
| cad_evo_forge | 11760 | 19,589,906 |
| longctx_forge | 30 | 49,735,767 |

## Pipeline
carry vetted v1.1 shards -> top-up fetch (rows+parquet bulk, offset-cursor resume) -> per-row gates (shape, non_latin, think-balance, dup, license, no_image) -> ONE global dedup index seeded with the carried corpus -> real gpt2 BPE per row -> per-slice pareto (strict) -> domain token-share caps -> deterministic interleave.

## Quota drops (rebalance)
| family | rows dropped | share before |
|---|---|---|
| reasoning_math | 79448 | 0.456 |

## Generator diversity (per source, post-pareto)
| source | distinct tasks | top task | top share |
|---|---|---|---|
| openmath_cot | 6 | math | 55% |
| opensci | 2 | science | 57% |
| openmath_tir | 4 | aops_c6_high_school_olympiads | 54% |
| openr1_math | 7 | olympiads | 73% |
| pyromind_short | 2 | tooluse | 57% |
| toolmind | 2 | tooluse | 54% |
| nemotron_swe | 2310 | swe | 8% |
| swesmith | 1685 | swe | 42% |
| coderforge | 801 | swe | 29% |
| ultradata_code | 3001 | coding | 3% |
| ultradata_general | 2001 | general | 8% |
| ultradata_search | 2001 | search | 0% |
| ollvm_re | 2 | re | 100% |
| vision_docmatix | 2 | vision | 88% |
| plot2code | 2 | vision | 97% |
| icons_forge | 9 | tabler-outline | 40% |
| re_synth_forge | 2937 | rle | 2% |
| cad_synth_forge | 23 | plate_holes | 10% |
| cad_evo_forge | 1258 | box_stack_pocket_boss | 12% |
| longctx_forge | 6 | utility library header rebuild | 37% |

Single-artifact families carry a concision envelope: at equal correctness
the pareto prune now drops bloated rows first, so the mixture teaches
tight, dense answers rather than padded ones.

## Files
- `train.jsonl.gz` — the mixture (interleaved; shuffle again before packing)
- `manifest.jsonl.gz` — per-row index: id, source, domain, turns, gpt2_tokens, tags
- `build/v1.8/shards/*.jsonl.gz` — per-source provenance pools
- `images.tar.gz` — PNGs referenced by `vision_image`/`vision_plot` rows (`meta.images`)

## Tag vocabulary (filter, never re-split)
`src:<source>` · `domain:<fine>` (agentic_general, agentic_search, agentic_swe, agentic_tool, cad_3d, code_agent, general_sft, reasoning_math, reasoning_sci, reasoning_science, reverse_eng, reverse_engineering, style_house, vision_image, vision_plot, vision_svg)
· `turns:<mono|short|mid|deep|mega>` · `tok:<1k|1k-5k|5k-20k|20k-50k|50k+>`
· `tools` + `toolcalls:<1-3|4-10|10+>` · `cot` · `multimodal` · `verified`

## Honest notes
- Carried rows keep their v1.0/v1.1 vetting; `meta.license` is per-row.
- Carry-only sources: agenticsft/demyagent (upstream license unverified),
  cadforge_3d (upstream dark since v1.0 — the 14 verified rows are all that exist).
- Vision rows' text token share is small by nature: the pixels live in
  `images.tar.gz`, outside any text token count. Every forge-rendered
  vision answer is program-verified upstream.
- Assistant turns carry `trainable: true` — mask non-assistant turns for loss.