HopChain v3.4: 7,900 verified multi-hop video questions (1080p captions, 397B gen)

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+captions/captions_1080p.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,156 @@
+---
+license: apache-2.0
+task_categories:
+  - video-text-to-text
+  - question-answering
+language:
+  - en
+tags:
+  - video
+  - multi-hop-reasoning
+  - rlvr
+  - verifiable-rewards
+  - reinforcement-learning
+size_categories:
+  - 1K<n<10K
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*.parquet
+---
+
+# HopChain v3.4 — multi-hop video reasoning with verifiable answers (1080p)
+
+**7,900 multi-hop video questions** over **1,730 videos**, each with an **integer ground-truth answer**
+that is checkable by rule (no LLM judge at train time). Built for **RLVR** training.
+
+Each question chains several **conditional-predicate hops**. A hop observes one robust categorical fact
+about a scene (existence / colour / category / action / setting) and a binary rule turns it into a number —
+`if <fact> then A else B` — and the numbers are aggregated with `+ - *` into a single verifiable integer.
+Because *which* branch fires is never stated, the question cannot be answered without watching the video.
+
+---
+
+## ⚠ Read this before using the data
+
+**1. The corpus is a MIXED caption oracle.** 144 of 1,818 videos (7.9%) were captioned by
+`Qwen3.5-122B-A10B`; the other 1,674 by `Qwen3.5-397B-A17B-FP8`. This was forced, not chosen: the 397B has
+32 attention heads, so vLLM accepts only TP ∈ {2,4,8}, and at TP=4 its 406 GB of weights need 101.5 GB/GPU
+— over an 80 GB card. One node in the cluster had 7 healthy GPUs and could not host it in any configuration.
+**The captioner is stamped per video in `_caption_meta.model`** (see `captions/captions_1080p.jsonl`), so
+the subset is identifiable. **Filter or audit it before any cross-video comparison.**
+
+**2. Generation never sees a frame.** The caption is the pipeline's *only* ground truth. A caption that
+fabricates poisons every question built on it. That is why resolution mattered so much (below) — but it also
+means **residual caption errors are the dominant remaining source of label noise.**
+
+**3. The graders were sampled (temperature 0.7), not greedy.** The keep/drop verdict is reproducible
+(per-request seeds) but not deterministic-by-design.
+
+---
+
+## The headline finding: 360p was the ceiling, not the captioner
+
+Earlier work on this corpus concluded that a 397B captioner *fabricated more on-screen text* than a smaller
+122B. **That conclusion was an artifact of resolution.** At 640×360, distant signs, watermarks and brand
+text are physically unresolvable — every model was guessing at a blur.
+
+Clean A/B — **same model, same prompt, same segment boundaries; resolution the only variable** — on 9
+concrete on-screen-text probes (word-boundary matched):
+
+| Qwen3.5-397B-A17B-FP8 | reads correctly | **misreads** | stays silent |
+|---|---|---|---|
+| @ 360p | 1 | **6** | 2 |
+| @ **1080p** | **4** | **1** | 4 |
+
+Fixed by resolution alone: `Southern Miss` → **Southern Utah**; `Cornstock` → **Comstock**;
+`SLOW AREA` → **REST AREA**. A fabricated `MIAMI` and a phantom HUD number `0397` both **vanish** — at 1080p
+the model goes **silent** rather than inventing, which is the failure mode that actually matters.
+
+---
+
+## The metric this dataset optimises: `blind_solver_rate`
+
+`blind_solver_rate` = accuracy of a solver that **never watches the video** and always takes the same
+branch. It is the number to keep *low*.
+
+| version | `then_rate` (healthy 0.35–0.65) | **`blind_solver_rate`** | keepers/video |
+|---|---|---|---|
+| v3.1 (360p) | 0.939 | **0.839** | 3.47 |
+| v3.2 (360p) | 0.627 | **0.261** | 1.76 |
+| **v3.4 (this release, 1080p)** | **0.507** | **0.133** | **4.47** |
+
+`then_rate` 0.507 is a near-perfect 50/50 split (all-THEN 0.133 vs all-ELSE 0.126) → **no branch bias**.
+
+**Answer diversity:** 964 distinct answers; the modal answer covers only ~1.3% of rows, so a
+constant-guess baseline scores ~1.3%. The 13.3% blind-solver figure comes from *branch strategy*, not from
+answers being guessable.
+
+### ⚠ Do NOT credit resolution for the yield gains
+
+Judge survival rose 30.8% → 59.6% between v3.2 and v3.4, but **six things changed** (caption source,
+generator 122B→397B, 6→8 candidates/video, greedy→sampled decoding, a batched-rewrite refactor, corpus
+size). Per-candidate drop rates:
+
+| drop code | v3.2 (360p) | v3.4 (1080p) | Δ | mostly attributable to |
+|---|---|---|---|---|
+| `wrong_branch` | 52.7% | 23.4% | −29.4pp | generator strength + fix-loop |
+| `answer_mismatch` | 54.6% | 25.4% | −29.2pp | generator strength + fix-loop |
+| `not_stated` | 8.8% | 2.7% | −6.1pp | **caption fidelity** |
+| `unfaithful_quote` | 18.0% | 15.0% | −3.1pp | **caption fidelity** |
+| `attachment_error` | 0.3% | 1.5% | **+1.2pp** | ← counter-signal |
+| **caption-fidelity codes combined** | 27.1% | **19.1%** | **−8.0pp** | **resolution** |
+
+The headline −29pp is **generator strength, not resolution**. Resolution's isolated contribution is the
+**−8.0pp** on caption-fidelity codes. `attachment_error` *rose*: richer captions describe more objects, so
+there are more chances to mis-attach an attribute.
+
+---
+
+## Contents
+
+| path | what |
+|---|---|
+| `data/train-*.parquet` | **7,900 keepers.** verl RLVR schema: `prompt`, `videos`, `reward_model` (`ground_truth`), `extra_info`, plus flattened `answer` and `video_id`. |
+| `captions/captions_1080p.jsonl` | 1,818 videos × segment-level captions — **the oracle**. Carries `_caption_meta` (model, max_pixels, fps, max_frames), so the mixed-oracle subset is auditable. |
+
+**Videos are not mirrored here.** They are already public at
+[`Choiszt/video_res`](https://huggingface.co/datasets/Choiszt/video_res) under `videos_1080p/`. Join on
+`video_id`.
+
+## Integrity (verified, 300-video sample / 1,331 rows)
+
+- ground truth empty: **0**
+- ground truth is a plain integer: **100%** — the verifiable-answer invariant
+- rows carrying a video reference: **100%**
+
+## How it was built
+
+`caption → gen → judge → too-easy`, on 8× DGX (H100-80GB).
+
+| stage | model | detail |
+|---|---|---|
+| caption | `Qwen3.5-397B-A17B-FP8` (TP=8) + `Qwen3.5-122B-A10B` (TP=4, 144 videos) | native 1080p (`max_pixels=2073600`, nothing downscaled), `fps=1`, ≤24 frames/segment |
+| gen + rewrite | `Qwen3.5-397B-A17B-FP8` | 8 candidates/video; rewrite assigns per-hop polarity, de-circularises scene refs, LLM fact-checks each condition against the caption |
+| judge | `Qwen3.6-27B` | the only semantic gate — arithmetic is evaluated **in code**, never by the model, and the judge is never shown the target answer |
+| too-easy | `Qwen3.5-9B` | solves each question 8× on the full video; drops it if solved every time |
+
+**Funnel:** 1,818 videos → 13,938 candidates → 8,313 judge survivors (59.6%) → **7,900 keepers**.
+51 videos (2.8%) yielded no candidates and are recorded, not silently dropped.
+
+Sampling (Qwen recipe): `temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, repetition_penalty=1.0`.
+`presence_penalty` is **split by output type** — `1.5` on prose captions (it fixes runaway repetition),
+`0.0` on every JSON stage (there it penalises re-emitting the very keys each hop requires, and malformed
+JSON silently costs candidates).
+
+## Known gaps
+
+- 53 of the original 1,871 videos have no 1080p source → excluded (a 360p subset would reintroduce a
+  mixed-**resolution** oracle).
+- 51 videos yielded 0 candidates (2.8%) — ordinary generation/parse misses, *not* context overflow
+  (verified: their caption-length distribution matches the successes).
+
+## Citation
+
+Derived from the video corpus of [`Choiszt/video_res`](https://huggingface.co/datasets/Choiszt/video_res).

captions/captions_1080p.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd3eee61f8d6b5a294283e4c44acb07646754680a3c308a285d3a9a69299440a
+size 115176476

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3f55f949af9078244f1bca0de52a12efcb3bbfd629c652bd083b916540351f6
+size 2272736