Upload folder using huggingface_hub

.gitattributes

@@ -58,3 +58,6 @@ 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
+provenance/mask-flags-v1.jsonl filter=lfs diff=lfs merge=lfs -text
+raw/v0_9_final.jsonl filter=lfs diff=lfs merge=lfs -text
+superseded-correct-policy/v0_9_final.correct-policy.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,148 @@
+# **Qwen3.5-35B-A3B — v0.9 dataset**
+
+Last substantive update: 2026-08-04
+
+**v0.9 is v0.7 with 38,145 assistant turns excluded from the loss. Nothing else
+differs — same rows, same order, same content, same reasoning, same pack seed.**
+
+The design and its evidence are in `QWEN35_V0_9_DATASET_PLAN.md`. This file records
+what was actually built, where it is, and how to check it.
+
+---
+
+## **Location — b200 cluster**
+
+    /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/
+      raw/v0_9_final.jsonl              742,665 rows, 34,528,993,378 bytes
+      packed-65k/train/                 137,384 packs in 8 parquet shards
+      provenance/mask-flags-v1.jsonl    38,145 records, one per masked turn
+      provenance/v0_9_build_report.json
+      provenance/v0_9_verify_report.json
+      provenance/census_v0_9_v2.json
+      provenance/*.py, *.sbatch         the exact scripts that produced it
+      logs/
+
+| artifact | SHA-256 |
+|---|---|
+| base `v0.7/raw/v0_7_final.jsonl` | `cd94b138f7988b81c48e12ff3b75559192ef7c4b48fdc2f86d076e48d830266d` |
+| **`raw/v0_9_final.jsonl`** | `7f742b2b3c3ab569c007070ca7d17089cc03d1cf4fa9375cccd58dc305b62a2f` |
+| staged v0.8 classifier flags | `a5d2f1f5999e7f03b495ff3e389656574c89a4e2830cb1e7bac3d0777299a0c1` |
+
+Superseded artifacts from the abandoned `--claim-policy correct` build are kept under
+`superseded-correct-policy/`; they are not inputs to anything.
+
+---
+
+## **What is masked**
+
+38,145 turns across 33,291 rows — **5.1% of assistant turns in 4.5% of rows**. Reasons
+overlap (most stalls are also malformed JSON), so each masked turn carries the set of
+reasons it qualifies for. The per-reason totals sum to 66,420; the union is 38,145.
+
+- **32,836 `terminus-json`** — the Terminus-2 harness rejected the action object
+- **28,292 `stall`** — empty content (28,270), no `commands` key (20), empty mini-swe
+  content (2)
+- **3,116 `false-claim`** — final assistant turn asserting `task_complete: true` on a row
+  a trusted verifier failed (numeric non-bool reward, explicit `passed`, no infra
+  exception, not a known-broken verifier family). 84 further candidates also carried an
+  executable command and were left alone
+- **1,385 `toolcall-syntax`** — malformed or schema-invalid tool call; 484 land on the
+  general-agent sources τ³ measures (API-Bank 478, Toucan 5, ToolACE 1)
+- **791 `prose-command`** — mini-swe turns that write a command as prose instead of
+  executing it (727 of 738 `no-toolcall`, 64 of 305 `trailing-after-submit`)
+
+Combination counts: `stall+terminus-json` 28,275 · `terminus-json` alone 4,561 ·
+`false-claim` 3,116 · `toolcall-syntax` 1,385 · `prose-command` 791 · `stall` alone 17.
+
+## **What is deliberately left trained**
+
+- **9,692 `rewrites`** — v0.7 has no masks and this is the most error-reactive class
+  (58.66%), while τ³ error recovery is the known weakness
+- **252 prose-only mini-swe turns** — end-of-run summaries with no command text; v0.7
+  trained them normally
+- **200 `terminus-unparsable-harness-rejected`** turns that carry no `terminus-json`
+  class. The other 4,264 unparsable turns do carry it and are masked
+- **796 rows that passed** contain 819 masked turns. This is intended: a stall inside a
+  trajectory that eventually passed is still a stall
+
+---
+
+## **How it was built**
+
+    workflows/sft/v0_9/census_v0_9.py     measure only, never mutates
+    workflows/sft/v0_9/build_v0_9.py      one streaming pass over the base
+    workflows/sft/v0_9/verify_v0_9.py     independent on-disk re-check
+    workflows/sft/v0_9/build_v0_9.sbatch  b200 launcher (1:56 wall)
+    workflows/sft/v0_9/pack_v0_9.sbatch   b200 pack at 65,536
+
+Packed with the same artifacts that produced v0.7 and v0.8 on b300 — packer
+`976ce4bb…`, `tokenizer.json` `06b95093…`, `chat_template.jinja` `f8a27a12…`, staged at
+`agentic-rl-260803/env/qwen35-qwen36-thinking-generation-spans`. A 300-row sample
+tokenizes to an identical digest on both clusters
+(`1594220b35a637c8…`; transformers 5.8.1, tokenizers 0.22.2, megatron-bridge 0.5.0 on
+each), so the b200 pack is not a different rendering of the corpus.
+
+**Seed 20260731 and 8 shards, matching v0.7**, on v0.7's row order — shard membership is
+identical, so a v0.7↔v0.9 comparison is not confounded by a different packing.
+
+## **Pack result, against v0.7**
+
+| | v0.7 | **v0.9** | delta |
+|---|---:|---:|---:|
+| rows in | 742,665 | 742,665 | 0 |
+| rows packed | 742,661 | **742,656** | **−5** |
+| packs | 137,387 | **137,384** | −3 |
+| tokens | 8,978,973,597 | **8,978,945,130** | −28,467 |
+| packing efficiency | 99.71% | **99.73%** | — |
+| rows over 65,536 tokens | 4 | **4** | 0 |
+| one epoch @ GBS16 | — | **8,587 steps** | — |
+
+**The whole difference is 5 rows, and they are accounted for.** `rows_over_max_len` is 4
+in both, which independently confirms no content grew or shrank; the extra 5 dropped rows
+are `rows_fully_masked` — trajectories in which *every* assistant turn was a defect, so
+nothing trainable survived. The packer drops those rather than emit a zero-loss sample.
+
+Loss coverage: **54,577,131 masked tokens against 3,408,327,232 trainable** — masking
+removes **1.58%** of the assistant-target loss. Span integrity was clean:
+`anomalies=0`, `turn_flags_on_non_assistant=0`, `span_turn_count_mismatch=0`,
+`span_bad_start=0`, `span_multiple_headers=0`, so every mask landed on the token span it
+was meant to. The validator additionally sampled 128 packs and read back 3,001,541
+assistant loss tokens.
+
+## **Gates that passed**
+
+Build (`v0_9_build_report.json`):
+
+- base and flag digests re-checked at build time; output digest recorded
+- 742,665 rows in, 742,665 out; no rows added, dropped or reordered
+- every modified row compared against source-plus-declared-changes — an unintended edit
+  anywhere is a nonzero exit; 0 occurred
+- 0 alignment anomalies: every flag resolved to an assistant turn
+- per-reason counts equal to the census figures, exactly
+- 4,963/4,963 untouched-row probes re-serialise byte-identically
+
+Verify (`v0_9_verify_report.json`), separate logic reading both files off disk:
+
+- **0 content corrections seen** — the only field that changed anywhere is
+  `step_loss_mask`
+- the base carried no `step_loss_mask` of its own, so all 38,145 are this build's
+- every masked turn's source content still matches the fingerprint recorded at build time
+- every `false-claim` mask is the final assistant turn of a `passed == False` row whose
+  parsed object carried `task_complete: true`
+- 0 errors
+
+Pack (`packed-65k/validation-summary.json`), `validate_qwen35_packed_dataset.py`:
+
+- 742,656 of 742,665 samples present, 9 dropped and both causes accounted for
+- 137,384 packs at 65,536 tokens, 99.7262% efficiency
+- 0 span anomalies of any kind; masks land on the intended token spans
+
+## **Training**
+
+3 epochs, TP2/EP8, MBS1/GBS16, LR 1e-5/1e-6, save every 500 steps. One epoch is **8,587
+steps** (137,384 packs / GBS 16) — the same as v0.7, whose 137,387 packs also round to
+8,587 — so matched-step comparisons against v0.7 are exact. Two epochs is **iter 17,174**
+and three is **25,761**; with saves every 500, the nearest checkpoints are 17,000 and
+25,500. Point `packed_data_path` at `datasets/v0.9/packed-65k`. Evaluation contract,
+hard floors and the checkpoint-selection rule are predeclared in
+`QWEN35_V0_9_DATASET_PLAN.md` §5 — read it before scoring, not after.

logs/build-23252.log

@@ -0,0 +1,80 @@
+v0.9 build started=2026-08-04T06:54:40+00:00 host=three-1
+source sha256 cd94b138f7988b81c48e12ff3b75559192ef7c4b48fdc2f86d076e48d830266d
+flags sha256 a5d2f1f5999e7f03b495ff3e389656574c89a4e2830cb1e7bac3d0777299a0c1
+flags in: 45,173 over 38,018 rows
+  50,000 rows  masked=2,479  corrected=215
+  100,000 rows  masked=4,984  corrected=435
+  150,000 rows  masked=7,496  corrected=672
+  200,000 rows  masked=10,110  corrected=897
+  250,000 rows  masked=12,608  corrected=1,099
+  300,000 rows  masked=15,146  corrected=1,316
+  350,000 rows  masked=17,677  corrected=1,546
+  400,000 rows  masked=20,184  corrected=1,801
+  450,000 rows  masked=22,775  corrected=2,017
+  500,000 rows  masked=25,348  corrected=2,276
+  550,000 rows  masked=27,909  corrected=2,493
+  600,000 rows  masked=30,461  corrected=2,723
+  650,000 rows  masked=33,023  corrected=2,947
+  700,000 rows  masked=34,979  corrected=3,116
+{
+ "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.7/raw/v0_7_final.jsonl",
+ "src_sha256": "cd94b138f7988b81c48e12ff3b75559192ef7c4b48fdc2f86d076e48d830266d",
+ "flags_in_file": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1-from-v0_8.jsonl",
+ "flags_in_sha256": "a5d2f1f5999e7f03b495ff3e389656574c89a4e2830cb1e7bac3d0777299a0c1",
+ "out": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+ "out_sha256": "20ea0a8d71f1c1bf22599082fd67f9feb34195cf8e32d375f128d01c401425aa",
+ "out_flags": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1.jsonl",
+ "rows": 742665,
+ "rows_modified": 33291,
+ "rows_with_masked_turns": 30226,
+ "rows_corrected": 3116,
+ "turns_masked": 35029,
+ "corrections": 3116,
+ "mask_labels_overlapping": {
+  "prose-command": 791,
+  "stall": 28292,
+  "terminus-json": 32836,
+  "toolcall-syntax": 1385
+ },
+ "mask_label_combinations": {
+  "stall+terminus-json": 28275,
+  "terminus-json": 4561,
+  "toolcall-syntax": 1385,
+  "prose-command": 791,
+  "stall": 17
+ },
+ "v0_8_flags_left_trained_by_class": {
+  "missing-action": 452,
+  "rewrites": 9692
+ },
+ "claims_by_tier": {
+  "infra|passed=False": 22,
+  "infra|passed=True": 1,
+  "trusted|passed=False": 3200,
+  "trusted|passed=True": 1750,
+  "unverified|passed=False": 1475,
+  "unverified|passed=None": 155532
+ },
+ "correction_status": {
+  "excluded:claim_plus_action": 84,
+  "ok": 3116
+ },
+ "correction_sources": {
+  "nvidia/Nemotron-Terminal-Synthetic-Tasks": 2146,
+  "open-thoughts/TaskTrove": 970
+ },
+ "masked_rows_with_passed_true": 796,
+ "masked_turns_in_passed_true_rows": 819,
+ "turns_both_masked_and_corrected": 0,
+ "roundtrip_probe": {
+  "tested": 4963,
+  "byte_identical": 4963
+ },
+ "anomalies": 0,
+ "anomaly_sample": [],
+ "limit": null,
+ "packed": false,
+ "shuffled": false
+}
+ALL GATES PASSED
+v0.9 build complete=2026-08-04T06:56:34+00:00

logs/build-23260.log

@@ -0,0 +1,83 @@
+v0.9 build started=2026-08-04T07:05:09+00:00 host=three-1
+source sha256 cd94b138f7988b81c48e12ff3b75559192ef7c4b48fdc2f86d076e48d830266d
+flags sha256 a5d2f1f5999e7f03b495ff3e389656574c89a4e2830cb1e7bac3d0777299a0c1
+flags in: 45,173 over 38,018 rows
+  50,000 rows  masked=2,694  claims=215
+  100,000 rows  masked=5,419  claims=435
+  150,000 rows  masked=8,168  claims=672
+  200,000 rows  masked=11,007  claims=897
+  250,000 rows  masked=13,707  claims=1,099
+  300,000 rows  masked=16,462  claims=1,316
+  350,000 rows  masked=19,223  claims=1,546
+  400,000 rows  masked=21,985  claims=1,801
+  450,000 rows  masked=24,792  claims=2,017
+  500,000 rows  masked=27,624  claims=2,276
+  550,000 rows  masked=30,402  claims=2,493
+  600,000 rows  masked=33,184  claims=2,723
+  650,000 rows  masked=35,970  claims=2,947
+  700,000 rows  masked=38,095  claims=3,116
+{
+ "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.7/raw/v0_7_final.jsonl",
+ "src_sha256": "cd94b138f7988b81c48e12ff3b75559192ef7c4b48fdc2f86d076e48d830266d",
+ "flags_in_file": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1-from-v0_8.jsonl",
+ "flags_in_sha256": "a5d2f1f5999e7f03b495ff3e389656574c89a4e2830cb1e7bac3d0777299a0c1",
+ "out": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+ "out_sha256": "7f742b2b3c3ab569c007070ca7d17089cc03d1cf4fa9375cccd58dc305b62a2f",
+ "out_flags": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1.jsonl",
+ "rows": 742665,
+ "rows_modified": 33291,
+ "rows_with_masked_turns": 33291,
+ "rows_content_corrected": 0,
+ "turns_masked": 38145,
+ "claim_policy": "mask",
+ "false_claims_suppressed": 3116,
+ "mask_labels_overlapping": {
+  "false-claim": 3116,
+  "prose-command": 791,
+  "stall": 28292,
+  "terminus-json": 32836,
+  "toolcall-syntax": 1385
+ },
+ "mask_label_combinations": {
+  "stall+terminus-json": 28275,
+  "terminus-json": 4561,
+  "false-claim": 3116,
+  "toolcall-syntax": 1385,
+  "prose-command": 791,
+  "stall": 17
+ },
+ "v0_8_flags_left_trained_by_class": {
+  "missing-action": 452,
+  "rewrites": 9692
+ },
+ "claims_by_tier": {
+  "infra|passed=False": 22,
+  "infra|passed=True": 1,
+  "trusted|passed=False": 3200,
+  "trusted|passed=True": 1750,
+  "unverified|passed=False": 1475,
+  "unverified|passed=None": 155532
+ },
+ "claim_status": {
+  "excluded:claim_plus_action": 84,
+  "ok": 3116
+ },
+ "claim_sources": {
+  "nvidia/Nemotron-Terminal-Synthetic-Tasks": 2146,
+  "open-thoughts/TaskTrove": 970
+ },
+ "masked_rows_with_passed_true": 796,
+ "masked_turns_in_passed_true_rows": 819,
+ "false_claim_turns_already_masked": 0,
+ "roundtrip_probe": {
+  "tested": 4963,
+  "byte_identical": 4963
+ },
+ "anomalies": 0,
+ "anomaly_sample": [],
+ "limit": null,
+ "packed": false,
+ "shuffled": false
+}
+ALL GATES PASSED
+v0.9 build complete=2026-08-04T07:07:04+00:00

logs/pack-23262.log

@@ -0,0 +1,90 @@
+empty reasoning template check passed
+packing rows=742665 src=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl out=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k seed=20260731 started=2026-08-04T07:08:45+00:00
+[pretokenize] tokenizer=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/qwen35-qwen36-thinking-generation-spans eos_id=248044 max_len=65536 pack=65536 workers=56 shards=8 rank=0/1 host=three-1
+  [tok] shard 0: 20032 kept (<= 65536 tok)
+  [tok] shard 0: 40000 kept (<= 65536 tok)
+  [tok] shard 0: 60031 kept (<= 65536 tok)
+  [tok] shard 0: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.71% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.379 <<<<<
+[pack] train_000: 92832 samples → 17258 packs (65536 tok each), 1127.8M tokens, packing_metadata={'dataset_max_seqlen': 58722, 'max_samples_per_bin': 12, 'packing_factor': 5.38, 'packing_efficiency': 99.71, 'pack_size': 65536, 'min_packed_seqlen': 30980} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_000.parquet
+  [tok] shard 1: 20032 kept (<= 65536 tok)
+  [tok] shard 1: 40000 kept (<= 65536 tok)
+  [tok] shard 1: 60032 kept (<= 65536 tok)
+  [tok] shard 1: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.71% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.417 <<<<<
+[pack] train_001: 92832 samples → 17136 packs (65536 tok each), 1119.9M tokens, packing_metadata={'dataset_max_seqlen': 63648, 'max_samples_per_bin': 12, 'packing_factor': 5.42, 'packing_efficiency': 99.71, 'pack_size': 65536, 'min_packed_seqlen': 42264} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_001.parquet
+  [tok] shard 2: 20032 kept (<= 65536 tok)
+  [tok] shard 2: 40000 kept (<= 65536 tok)
+  [tok] shard 2: 60032 kept (<= 65536 tok)
+  [tok] shard 2: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.71% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.390 <<<<<
+[pack] train_002: 92833 samples → 17224 packs (65536 tok each), 1125.7M tokens, packing_metadata={'dataset_max_seqlen': 61151, 'max_samples_per_bin': 12, 'packing_factor': 5.39, 'packing_efficiency': 99.71, 'pack_size': 65536, 'min_packed_seqlen': 59514} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_002.parquet
+  [tok] shard 3: 20032 kept (<= 65536 tok)
+  [tok] shard 3: 40000 kept (<= 65536 tok)
+  [tok] shard 3: 60030 kept (<= 65536 tok)
+  [tok] shard 3: 80062 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.72% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.395 <<<<<
+[pack] train_003: 92831 samples → 17208 packs (65536 tok each), 1124.7M tokens, packing_metadata={'dataset_max_seqlen': 60748, 'max_samples_per_bin': 13, 'packing_factor': 5.39, 'packing_efficiency': 99.72, 'pack_size': 65536, 'min_packed_seqlen': 28348} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_003.parquet
+  [tok] shard 4: 20032 kept (<= 65536 tok)
+  [tok] shard 4: 40000 kept (<= 65536 tok)
+  [tok] shard 4: 60032 kept (<= 65536 tok)
+  [tok] shard 4: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.72% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.409 <<<<<
+[pack] train_004: 92833 samples → 17164 packs (65536 tok each), 1121.8M tokens, packing_metadata={'dataset_max_seqlen': 61299, 'max_samples_per_bin': 13, 'packing_factor': 5.41, 'packing_efficiency': 99.72, 'pack_size': 65536, 'min_packed_seqlen': 44958} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_004.parquet
+  [tok] shard 5: 20032 kept (<= 65536 tok)
+  [tok] shard 5: 40063 kept (<= 65536 tok)
+  [tok] shard 5: 60031 kept (<= 65536 tok)
+  [tok] shard 5: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.392 <<<<<
+[pack] train_005: 92831 samples → 17218 packs (65536 tok each), 1125.4M tokens, packing_metadata={'dataset_max_seqlen': 61168, 'max_samples_per_bin': 14, 'packing_factor': 5.39, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 23337} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_005.parquet
+  [tok] shard 6: 20031 kept (<= 65536 tok)
+  [tok] shard 6: 40063 kept (<= 65536 tok)
+  [tok] shard 6: 60031 kept (<= 65536 tok)
+  [tok] shard 6: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.437 <<<<<
+[pack] train_006: 92832 samples → 17075 packs (65536 tok each), 1116.1M tokens, packing_metadata={'dataset_max_seqlen': 63891, 'max_samples_per_bin': 13, 'packing_factor': 5.44, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 47238} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_006.parquet
+  [tok] shard 7: 20032 kept (<= 65536 tok)
+  [tok] shard 7: 40000 kept (<= 65536 tok)
+  [tok] shard 7: 60032 kept (<= 65536 tok)
+  [tok] shard 7: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.71% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.428 <<<<<
+[pack] train_007: 92832 samples → 17101 packs (65536 tok each), 1117.6M tokens, packing_metadata={'dataset_max_seqlen': 56757, 'max_samples_per_bin': 14, 'packing_factor': 5.43, 'packing_efficiency': 99.71, 'pack_size': 65536, 'min_packed_seqlen': 46447} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train/train_007.parquet
+[pretokenize] DONE → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k  (8.98B train tokens across 137384 packs, 8 shard(s))
+[turn-mask] rows_with_masked_turns=33291 masked_turns=38145 masked_tokens=54577131 trainable_tokens=3408327232 rows_fully_masked=5 anomalies=0 turn_flags_on_non_assistant=0 span_turn_count_mismatch=0 span_bad_start=0 span_multiple_headers=0
+{
+  "dropped_overlength_samples": 9,
+  "effective_tokens": 8978945130,
+  "one_epoch_steps_gbs16": 8587,
+  "one_epoch_steps_gbs32": 4294,
+  "packing_efficiency_percent": 99.7262,
+  "packs": 137384,
+  "sampled_assistant_loss_tokens": 3001541,
+  "sampled_packs_checked_for_loss_masks": 128,
+  "samples": 742656,
+  "source_samples": 742665
+}
+packing complete at=2026-08-04T08:20:47+00:00

logs/pack-23521.log

@@ -0,0 +1,90 @@
+empty reasoning template check passed ('<think>\n\n</think>')
+packing rows=742665 src=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl out=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b seed=20260731 started=2026-08-04T22:12:30+00:00
+[pretokenize] tokenizer=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/qwen3-30b-a3b-thinking-generation-spans eos_id=151643 max_len=65536 pack=65536 workers=56 shards=8 rank=0/1 host=three-20
+  [tok] shard 0: 20032 kept (<= 65536 tok)
+  [tok] shard 0: 40000 kept (<= 65536 tok)
+  [tok] shard 0: 60031 kept (<= 65536 tok)
+  [tok] shard 0: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.593 <<<<<
+[pack] train_000: 92833 samples → 16598 packs (65536 tok each), 1084.9M tokens, packing_metadata={'dataset_max_seqlen': 64906, 'max_samples_per_bin': 13, 'packing_factor': 5.59, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 29831} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_000.parquet
+  [tok] shard 1: 20032 kept (<= 65536 tok)
+  [tok] shard 1: 40000 kept (<= 65536 tok)
+  [tok] shard 1: 60032 kept (<= 65536 tok)
+  [tok] shard 1: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.632 <<<<<
+[pack] train_001: 92832 samples → 16483 packs (65536 tok each), 1077.4M tokens, packing_metadata={'dataset_max_seqlen': 61011, 'max_samples_per_bin': 13, 'packing_factor': 5.63, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 17157} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_001.parquet
+  [tok] shard 2: 20032 kept (<= 65536 tok)
+  [tok] shard 2: 40000 kept (<= 65536 tok)
+  [tok] shard 2: 60032 kept (<= 65536 tok)
+  [tok] shard 2: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.605 <<<<<
+[pack] train_002: 92833 samples → 16564 packs (65536 tok each), 1082.7M tokens, packing_metadata={'dataset_max_seqlen': 59708, 'max_samples_per_bin': 14, 'packing_factor': 5.6, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 15580} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_002.parquet
+  [tok] shard 3: 20032 kept (<= 65536 tok)
+  [tok] shard 3: 40000 kept (<= 65536 tok)
+  [tok] shard 3: 60031 kept (<= 65536 tok)
+  [tok] shard 3: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.74% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.610 <<<<<
+[pack] train_003: 92832 samples → 16549 packs (65536 tok each), 1081.8M tokens, packing_metadata={'dataset_max_seqlen': 64963, 'max_samples_per_bin': 14, 'packing_factor': 5.61, 'packing_efficiency': 99.74, 'pack_size': 65536, 'min_packed_seqlen': 62753} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_003.parquet
+  [tok] shard 4: 20032 kept (<= 65536 tok)
+  [tok] shard 4: 40000 kept (<= 65536 tok)
+  [tok] shard 4: 60032 kept (<= 65536 tok)
+  [tok] shard 4: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.622 <<<<<
+[pack] train_004: 92833 samples → 16513 packs (65536 tok each), 1079.4M tokens, packing_metadata={'dataset_max_seqlen': 60128, 'max_samples_per_bin': 13, 'packing_factor': 5.62, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 23191} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_004.parquet
+  [tok] shard 5: 20032 kept (<= 65536 tok)
+  [tok] shard 5: 40063 kept (<= 65536 tok)
+  [tok] shard 5: 60031 kept (<= 65536 tok)
+  [tok] shard 5: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.74% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.605 <<<<<
+[pack] train_005: 92831 samples → 16562 packs (65536 tok each), 1082.7M tokens, packing_metadata={'dataset_max_seqlen': 59611, 'max_samples_per_bin': 12, 'packing_factor': 5.61, 'packing_efficiency': 99.74, 'pack_size': 65536, 'min_packed_seqlen': 58734} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_005.parquet
+  [tok] shard 6: 20031 kept (<= 65536 tok)
+  [tok] shard 6: 40063 kept (<= 65536 tok)
+  [tok] shard 6: 60031 kept (<= 65536 tok)
+  [tok] shard 6: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.74% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.652 <<<<<
+[pack] train_006: 92832 samples → 16424 packs (65536 tok each), 1073.7M tokens, packing_metadata={'dataset_max_seqlen': 62547, 'max_samples_per_bin': 13, 'packing_factor': 5.65, 'packing_efficiency': 99.74, 'pack_size': 65536, 'min_packed_seqlen': 57978} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_006.parquet
+  [tok] shard 7: 20032 kept (<= 65536 tok)
+  [tok] shard 7: 40000 kept (<= 65536 tok)
+  [tok] shard 7: 60032 kept (<= 65536 tok)
+  [tok] shard 7: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.73% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.643 <<<<<
+[pack] train_007: 92832 samples → 16452 packs (65536 tok each), 1075.4M tokens, packing_metadata={'dataset_max_seqlen': 54299, 'max_samples_per_bin': 14, 'packing_factor': 5.64, 'packing_efficiency': 99.73, 'pack_size': 65536, 'min_packed_seqlen': 36718} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train/train_007.parquet
+[pretokenize] DONE → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b  (8.64B train tokens across 132145 packs, 8 shard(s))
+[turn-mask] rows_with_masked_turns=33291 masked_turns=38145 masked_tokens=53754611 trainable_tokens=3343677840 rows_fully_masked=5 anomalies=0 turn_flags_on_non_assistant=0 span_turn_count_mismatch=0 span_bad_start=0 span_multiple_headers=0
+{
+  "dropped_overlength_samples": 7,
+  "effective_tokens": 8637920650,
+  "one_epoch_steps_gbs16": 8260,
+  "one_epoch_steps_gbs32": 4130,
+  "packing_efficiency_percent": 99.7421,
+  "packs": 132145,
+  "sampled_assistant_loss_tokens": 3269634,
+  "sampled_packs_checked_for_loss_masks": 128,
+  "samples": 742658,
+  "source_samples": 742665
+}
+packing complete at=2026-08-04T23:20:04+00:00

logs/pack-23522.log

@@ -0,0 +1,91 @@
+empty reasoning template check passed ('<think></think>')
+packing rows=742665 src=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl out=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano seed=20260731 started=2026-08-04T22:12:31+00:00
+[pretokenize] tokenizer=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/nemotron3-nano-30b-a3b-thinking-generation-spans eos_id=11 max_len=65536 pack=65536 workers=56 shards=8 rank=0/1 host=three-21
+  [tok] shard 0: 20032 kept (<= 65536 tok)
+  [tok] shard 0: 40000 kept (<= 65536 tok)
+  [tok] shard 0: 60031 kept (<= 65536 tok)
+  [tok] shard 0: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.69% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.291 <<<<<
+[pack] train_000: 92832 samples → 17544 packs (65536 tok each), 1146.3M tokens, packing_metadata={'dataset_max_seqlen': 61597, 'max_samples_per_bin': 13, 'packing_factor': 5.29, 'packing_efficiency': 99.69, 'pack_size': 65536, 'min_packed_seqlen': 6964} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_000.parquet
+  [tok] shard 1: 20032 kept (<= 65536 tok)
+  [tok] shard 1: 40000 kept (<= 65536 tok)
+  [tok] shard 1: 60032 kept (<= 65536 tok)
+  [tok] shard 1: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.69% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.328 <<<<<
+[pack] train_001: 92831 samples → 17422 packs (65536 tok each), 1138.4M tokens, packing_metadata={'dataset_max_seqlen': 63185, 'max_samples_per_bin': 14, 'packing_factor': 5.33, 'packing_efficiency': 99.69, 'pack_size': 65536, 'min_packed_seqlen': 42487} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_001.parquet
+  [tok] shard 2: 20031 kept (<= 65536 tok)
+  [tok] shard 2: 40063 kept (<= 65536 tok)
+  [tok] shard 2: 60031 kept (<= 65536 tok)
+  [tok] shard 2: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.69% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.302 <<<<<
+[pack] train_002: 92832 samples → 17509 packs (65536 tok each), 1144.0M tokens, packing_metadata={'dataset_max_seqlen': 63791, 'max_samples_per_bin': 11, 'packing_factor': 5.3, 'packing_efficiency': 99.69, 'pack_size': 65536, 'min_packed_seqlen': 33311} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_002.parquet
+  [tok] shard 3: 20032 kept (<= 65536 tok)
+  [tok] shard 3: 40000 kept (<= 65536 tok)
+  [tok] shard 3: 60030 kept (<= 65536 tok)
+  [tok] shard 3: 80062 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.70% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.306 <<<<<
+[pack] train_003: 92831 samples → 17495 packs (65536 tok each), 1143.2M tokens, packing_metadata={'dataset_max_seqlen': 63001, 'max_samples_per_bin': 12, 'packing_factor': 5.31, 'packing_efficiency': 99.7, 'pack_size': 65536, 'min_packed_seqlen': 28526} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_003.parquet
+  [tok] shard 4: 20032 kept (<= 65536 tok)
+  [tok] shard 4: 40000 kept (<= 65536 tok)
+  [tok] shard 4: 60032 kept (<= 65536 tok)
+  [tok] shard 4: 80000 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.69% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.320 <<<<<
+[pack] train_004: 92833 samples → 17450 packs (65536 tok each), 1140.1M tokens, packing_metadata={'dataset_max_seqlen': 64203, 'max_samples_per_bin': 12, 'packing_factor': 5.32, 'packing_efficiency': 99.69, 'pack_size': 65536, 'min_packed_seqlen': 24361} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_004.parquet
+  [tok] shard 5: 20032 kept (<= 65536 tok)
+  [tok] shard 5: 40000 kept (<= 65536 tok)
+  [tok] shard 5: 60031 kept (<= 65536 tok)
+  [tok] shard 5: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.70% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.304 <<<<<
+[pack] train_005: 92831 samples → 17503 packs (65536 tok each), 1143.7M tokens, packing_metadata={'dataset_max_seqlen': 61469, 'max_samples_per_bin': 13, 'packing_factor': 5.3, 'packing_efficiency': 99.7, 'pack_size': 65536, 'min_packed_seqlen': 47829} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_005.parquet
+  [tok] shard 6: 20031 kept (<= 65536 tok)
+  [tok] shard 6: 40063 kept (<= 65536 tok)
+  [tok] shard 6: 60031 kept (<= 65536 tok)
+  [tok] shard 6: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.70% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.347 <<<<<
+[pack] train_006: 92832 samples → 17362 packs (65536 tok each), 1134.6M tokens, packing_metadata={'dataset_max_seqlen': 65369, 'max_samples_per_bin': 13, 'packing_factor': 5.35, 'packing_efficiency': 99.7, 'pack_size': 65536, 'min_packed_seqlen': 40602} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_006.parquet
+  [tok] shard 7: 20032 kept (<= 65536 tok)
+  [tok] shard 7: 40000 kept (<= 65536 tok)
+  [tok] shard 7: 60032 kept (<= 65536 tok)
+  [tok] shard 7: 80000 kept (<= 65536 tok)
+  [tok] shard 7: 80063 kept (<= 65536 tok)
+INFO:megatron.bridge.data.datasets.packing_utils:Creating histogram from tokenized dataset...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing sequences to length 65536...
+INFO:megatron.bridge.data.datasets.packing_utils:Packing is 99.70% efficient
+INFO:megatron.bridge.data.datasets.packing_utils:>>>>> For pack size 65536, average number of sequences per pack is n = 5.339 <<<<<
+[pack] train_007: 92832 samples → 17386 packs (65536 tok each), 1136.0M tokens, packing_metadata={'dataset_max_seqlen': 57834, 'max_samples_per_bin': 13, 'packing_factor': 5.34, 'packing_efficiency': 99.7, 'pack_size': 65536, 'min_packed_seqlen': 24948} → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train/train_007.parquet
+[pretokenize] DONE → /NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano  (9.13B train tokens across 139671 packs, 8 shard(s))
+[turn-mask] rows_with_masked_turns=33291 masked_turns=38145 masked_tokens=55446770 trainable_tokens=3396608582 rows_fully_masked=5 anomalies=0 turn_flags_on_non_assistant=0 span_turn_count_mismatch=0 span_bad_start=0 span_multiple_headers=0
+{
+  "dropped_overlength_samples": 11,
+  "effective_tokens": 9126387297,
+  "one_epoch_steps_gbs16": 8730,
+  "one_epoch_steps_gbs32": 4365,
+  "packing_efficiency_percent": 99.704,
+  "packs": 139671,
+  "sampled_assistant_loss_tokens": 3054640,
+  "sampled_packs_checked_for_loss_masks": 128,
+  "samples": 742654,
+  "source_samples": 742665
+}
+packing complete at=2026-08-04T23:22:45+00:00

logs/v10-bucket-23465.log

@@ -0,0 +1,34 @@
+rows 742665
+== group ==
+  terminal: 406863 (54.8%)
+  general: 283655 (38.2%)
+  unknown: 52147 (7.0%)
+== masked turns by group ==
+  terminal: 37609
+  general: 484
+  unknown: 52
+== rendered_tokens by group (metadata rows only) ==
+  terminal: 6,383,791,721
+  general: 879,773,446
+== detail ==
+  bucket:agentic_tau2_bfcl: 168646
+  bucket:terminus2: 168073
+  bucket:non_terminus_terminal_swe: 167421
+  bucket:UNKNOWN: 52147
+  labeled:websearch: 43446
+  labeled:xlam-fc: 39490
+  labeled:apibank: 14258
+  labeled:terminus-2: 12277
+  labeled:pi-clean: 10858
+  labeled:pi: 10720
+  labeled:tau2-synth: 9603
+  labeled:mini-swe-agent-clean: 7893
+  labeled:mini-swe-agent-minimal: 7761
+  labeled:mini-swe-agent: 7054
+  labeled:toolace: 5059
+  labeled:opencode: 4979
+  labeled:opencode-clean: 3961
+  labeled:opencode-todo: 3141
+  labeled:long-context: 2993
+  labeled:pi-minimal: 2725
+  labeled:opencode-websearch: 160

logs/v10-src-23460.log

@@ -0,0 +1,150 @@
+rows 742665
+== top-level keys ==
+  tools: 690518
+  metadata: 556287
+  agent: 186378
+  api_exchange_count: 186378
+  api_provider: 186378
+  api_row_id: 186378
+  api_session_key: 186378
+  command_complexity: 186378
+  completion_tokens: 186378
+  container_def: 186378
+  date: 186378
+  description: 186378
+  domain: 186378
+  exception_message: 186378
+  exception_type: 186378
+  harness: 186378
+  language: 186378
+  model: 186378
+  model_provider: 186378
+  outcome: 186378
+  parent_row_id: 186378
+  passed: 186378
+  primitive_skills: 186378
+  prompt_tokens: 186378
+  provider_metadata: 186378
+  reasoning_complete: 186378
+  result_path: 186378
+  reward: 186378
+  root_row_id: 186378
+  run_id: 186378
+  scenario: 186378
+  segment_index: 186378
+  segment_reason: 186378
+  skill_type: 186378
+  source: 186378
+  source_archive: 186378
+  source_path: 186378
+  task: 186378
+  task_complexity: 186378
+  task_id: 186378
+== source ==
+  None: 556287
+  nvidia/Nemotron-Terminal-Synthetic-Tasks: 43386
+  Salesforce/xlam-function-calling-60k: 38877
+  xanhho/2WikiMultihopQA: 28844
+  open-thoughts/TaskTrove: 27983
+  hotpotqa/hotpot_qa: 14602
+  liminghao1630/API-Bank: 14258
+  Team-ACE/ToolACE (seeds): 5059
+  synth-fresh-db/airline: 3269
+  longtool:liminghao1630/API-Bank: 2812
+  synth-fresh-db/retail: 2719
+  synth-fresh-db/home_connect: 1798
+  synth-fresh-db/med_scheduling: 690
+  Salesforce/xlam-function-calling-60k (augmented): 613
+  synth-fresh-db/cloud_billing_support: 568
+  synth-fresh-db/food_delivery_ops: 559
+  opencode-websearch/code: 160
+  longtool:Agent-Ark/Toucan-1.5M: 130
+  longtool:Team-ACE/ToolACE: 51
+== harness ==
+  None: 556287
+  websearch: 43446
+  xlam-fc: 39490
+  apibank: 14258
+  terminus-2: 12277
+  pi-clean: 10858
+  pi: 10720
+  tau2-synth: 9603
+  mini-swe-agent-clean: 7893
+  mini-swe-agent-minimal: 7761
+  mini-swe-agent: 7054
+  toolace: 5059
+  opencode: 4979
+  opencode-clean: 3961
+  opencode-todo: 3141
+  long-context: 2993
+  pi-minimal: 2725
+  opencode-websearch: 160
+== domain ==
+  None: 556287
+  websearch: 43446
+  xlam-fc: 39490
+  superuser: 15944
+  apibank: 14258
+  issue-tasks: 9145
+  debugging: 6389
+  security: 5928
+  file_operations: 5924
+  scientific_computing: 5914
+  software_engineering: 5880
+  system_administration: 5074
+  toolace: 5059
+  model_training: 4750
+  airline: 3269
+  long-context: 2993
+  retail: 2719
+  home_connect: 1798
+  swesmith: 1505
+  tezos: 1389
+  data_querying: 1262
+  dependency_management: 1197
+  data_processing: 1068
+  med_scheduling: 690
+  cloud_billing_support: 568
+  food_delivery_ops: 559
+  opencode-websearch: 160
+== source|harness|domain ==
+  ('None', 'None', 'None'): 556287
+  ('Salesforce/xlam-function-calling-60k', 'xlam-fc', 'xlam-fc'): 38877
+  ('xanhho/2WikiMultihopQA', 'websearch', 'websearch'): 28844
+  ('hotpotqa/hotpot_qa', 'websearch', 'websearch'): 14602
+  ('liminghao1630/API-Bank', 'apibank', 'apibank'): 14258
+  ('Team-ACE/ToolACE (seeds)', 'toolace', 'toolace'): 5059
+  ('synth-fresh-db/airline', 'tau2-synth', 'airline'): 3269
+  ('longtool:liminghao1630/API-Bank', 'long-context', 'long-context'): 2812
+  ('synth-fresh-db/retail', 'tau2-synth', 'retail'): 2719
+  ('open-thoughts/TaskTrove', 'opencode', 'superuser'): 2301
+  ('synth-fresh-db/home_connect', 'tau2-synth', 'home_connect'): 1798
+  ('open-thoughts/TaskTrove', 'terminus-2', 'issue-tasks'): 1779
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'terminus-2', 'debugging'): 1775
+  ('open-thoughts/TaskTrove', 'opencode-clean', 'superuser'): 1761
+  ('open-thoughts/TaskTrove', 'pi-clean', 'superuser'): 1752
+  ('open-thoughts/TaskTrove', 'terminus-2', 'superuser'): 1749
+  ('open-thoughts/TaskTrove', 'mini-swe-agent-clean', 'superuser'): 1701
+  ('open-thoughts/TaskTrove', 'mini-swe-agent-minimal', 'superuser'): 1675
+  ('open-thoughts/TaskTrove', 'pi', 'superuser'): 1590
+  ('open-thoughts/TaskTrove', 'opencode-todo', 'superuser'): 1478
+  ('open-thoughts/TaskTrove', 'mini-swe-agent', 'superuser'): 1362
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'security'): 1210
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'terminus-2', 'scientific_computing'): 1174
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi-clean', 'security'): 1156
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'terminus-2', 'system_administration'): 1141
+  ('open-thoughts/TaskTrove', 'opencode', 'issue-tasks'): 1121
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi-clean', 'software_engineering'): 1106
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'terminus-2', 'file_operations'): 1105
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi-clean', 'scientific_computing'): 1096
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi-clean', 'file_operations'): 1093
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'software_engineering'): 1090
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'file_operations'): 1059
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi-clean', 'debugging'): 1052
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'debugging'): 1024
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'scientific_computing'): 1018
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'model_training'): 945
+  ('open-thoughts/TaskTrove', 'opencode-clean', 'issue-tasks'): 942
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'terminus-2', 'software_engineering'): 918
+  ('open-thoughts/TaskTrove', 'pi', 'issue-tasks'): 909
+  ('nvidia/Nemotron-Terminal-Synthetic-Tasks', 'pi', 'system_administration'): 906

logs/verify-23257.log

@@ -0,0 +1,26 @@
+emitted flags: 35,029 over 30,226 rows
+  100,000 rows  errors=0
+  200,000 rows  errors=0
+  300,000 rows  errors=0
+  400,000 rows  errors=0
+  500,000 rows  errors=0
+  600,000 rows  errors=0
+  700,000 rows  errors=0
+{
+ "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.7/raw/v0_7_final.jsonl",
+ "out": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+ "flags": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1.jsonl",
+ "rows_src": 742665,
+ "rows_out": 742665,
+ "flags_emitted": 35029,
+ "masked_turns_seen": 35029,
+ "corrections_seen": 3116,
+ "correction_sources": {
+  "nvidia/Nemotron-Terminal-Synthetic-Tasks": 2146,
+  "open-thoughts/TaskTrove": 970
+ },
+ "correction_rewards_nonzero": 0,
+ "errors": 0,
+ "error_sample": []
+}
+VERIFY PASSED

logs/verify-23261.log

@@ -0,0 +1,30 @@
+emitted flags: 38,145 over 33,291 rows
+  100,000 rows  errors=0
+  200,000 rows  errors=0
+  300,000 rows  errors=0
+  400,000 rows  errors=0
+  500,000 rows  errors=0
+  600,000 rows  errors=0
+  700,000 rows  errors=0
+{
+ "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.7/raw/v0_7_final.jsonl",
+ "out": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+ "flags": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/mask-flags-v1.jsonl",
+ "rows_src": 742665,
+ "rows_out": 742665,
+ "flags_emitted": 38145,
+ "masked_turns_seen": 38145,
+ "masked_turns_by_label": {
+  "false-claim": 3116,
+  "prose-command": 791,
+  "stall": 28292,
+  "terminus-json": 32836,
+  "toolcall-syntax": 1385
+ },
+ "content_corrections_seen": 0,
+ "correction_sources": {},
+ "correction_rewards_nonzero": 0,
+ "errors": 0,
+ "error_sample": []
+}
+VERIFY PASSED

logs/vtc-v09-census-23725.out

@@ -0,0 +1,16 @@
+rows_scanned=50000
+rows_scanned=100000
+rows_scanned=150000
+rows_scanned=200000
+rows_scanned=250000
+rows_scanned=300000
+rows_scanned=350000
+rows_scanned=400000
+rows_scanned=450000
+rows_scanned=500000
+rows_scanned=550000
+rows_scanned=600000
+rows_scanned=650000
+rows_scanned=700000
+{"duplicate_first_user_rows": 61542, "malformed_rows": 0, "max_first_user_frequency": 12, "max_rendered_tokens": 32768, "rows": 742665, "rows_over_65536": 0, "unique_first_user": 442598}
+census_complete output=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/verify-then-claim-task-usage-v1 at=2026-08-05T06:50:42+00:00

logs/vtc-v09-census-23946.out

@@ -0,0 +1,16 @@
+rows_scanned=50000
+rows_scanned=100000
+rows_scanned=150000
+rows_scanned=200000
+rows_scanned=250000
+rows_scanned=300000
+rows_scanned=350000
+rows_scanned=400000
+rows_scanned=450000
+rows_scanned=500000
+rows_scanned=550000
+rows_scanned=600000
+rows_scanned=650000
+rows_scanned=700000
+{"duplicate_first_user_rows": 87645, "malformed_rows": 0, "max_first_user_frequency": 64, "max_rendered_tokens": 32768, "rows": 742665, "rows_over_65536": 0, "unique_first_user": 655020}
+census_complete output=/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/provenance/verify-then-claim-task-usage-v2 at=2026-08-05T16:37:48+00:00

packed-65k-nemotron3-nano/dataset_info.json

@@ -0,0 +1,32 @@
+{
+  "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+  "tokenizer": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/nemotron3-nano-30b-a3b-thinking-generation-spans",
+  "max_len": 65536,
+  "pack_size": 65536,
+  "shards": 8,
+  "kept_samples": 742654,
+  "train_samples": 742654,
+  "val_samples": 0,
+  "train_packs": 139671,
+  "val_packs": 0,
+  "train_tokens": 9126387297,
+  "val_tokens": 0,
+  "train_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/train",
+  "val_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/val.parquet",
+  "packed_metadata": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-nemotron3-nano/packed_metadata.json",
+  "turn_mask": {
+    "rows_seen": 742665,
+    "rows_kept": 742654,
+    "rows_over_max_len": 6,
+    "rows_no_trainable": 0,
+    "rows_with_masked_turns": 33291,
+    "rows_fully_masked": 5,
+    "masked_turns": 38145,
+    "masked_tokens": 55446770,
+    "trainable_tokens": 3396608582,
+    "turn_flags_on_non_assistant": 0,
+    "span_turn_count_mismatch": 0,
+    "span_bad_start": 0,
+    "span_multiple_headers": 0
+  }
+}

packed-65k-nemotron3-nano/packed_metadata.json

@@ -0,0 +1,66 @@
+[
+  {
+    "dataset_max_seqlen": 61597,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.29,
+    "packing_efficiency": 99.69,
+    "pack_size": 65536,
+    "min_packed_seqlen": 6964
+  },
+  {
+    "dataset_max_seqlen": 63185,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.33,
+    "packing_efficiency": 99.69,
+    "pack_size": 65536,
+    "min_packed_seqlen": 42487
+  },
+  {
+    "dataset_max_seqlen": 63791,
+    "max_samples_per_bin": 11,
+    "packing_factor": 5.3,
+    "packing_efficiency": 99.69,
+    "pack_size": 65536,
+    "min_packed_seqlen": 33311
+  },
+  {
+    "dataset_max_seqlen": 63001,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.31,
+    "packing_efficiency": 99.7,
+    "pack_size": 65536,
+    "min_packed_seqlen": 28526
+  },
+  {
+    "dataset_max_seqlen": 64203,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.32,
+    "packing_efficiency": 99.69,
+    "pack_size": 65536,
+    "min_packed_seqlen": 24361
+  },
+  {
+    "dataset_max_seqlen": 61469,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.3,
+    "packing_efficiency": 99.7,
+    "pack_size": 65536,
+    "min_packed_seqlen": 47829
+  },
+  {
+    "dataset_max_seqlen": 65369,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.35,
+    "packing_efficiency": 99.7,
+    "pack_size": 65536,
+    "min_packed_seqlen": 40602
+  },
+  {
+    "dataset_max_seqlen": 57834,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.34,
+    "packing_efficiency": 99.7,
+    "pack_size": 65536,
+    "min_packed_seqlen": 24948
+  }
+]

packed-65k-nemotron3-nano/train/train_000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee38edb6e6c33a36b631edac0ef98747632ddc6ed8dfd81a181064ced96eb8ae
+size 1252131644

packed-65k-nemotron3-nano/train/train_001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edb19157e2006e2ae6e5422bf9682f4fac6b337c1bb332dca9a49e0c00e925e6
+size 1241821185

packed-65k-nemotron3-nano/train/train_002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37c5416eb7297f1c49fb9f2fef27e09c10f98c8faadde3c3c4dcdea3048f26f1
+size 1247957612

packed-65k-nemotron3-nano/train/train_003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c84e13ea84008eb9ce6eddfc4cb5b1957cc563701fe9f40e0223b32197d45f1
+size 1244839111

packed-65k-nemotron3-nano/train/train_004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d892c1947b484cfb7e7e4d8b485767c4315471c28949065d40d1dd854c755e9b
+size 1243565872

packed-65k-nemotron3-nano/train/train_005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:105ea5ba199995df1bb7e4171966d5679afdcb96be53f628bbc70df811134996
+size 1248467716

packed-65k-nemotron3-nano/train/train_006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b2be26bf962b3e023c3e20a0f425db734b1e5095a42eb0c754aea841f05e6d1
+size 1238205828

packed-65k-nemotron3-nano/train/train_007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7466b885ae1041977742f01198ff56b1b54862dbd6ea364e9052e682a9958d24
+size 1240990225

packed-65k-nemotron3-nano/validation-summary.json

@@ -0,0 +1,12 @@
+{
+  "dropped_overlength_samples": 11,
+  "effective_tokens": 9126387297,
+  "one_epoch_steps_gbs16": 8730,
+  "one_epoch_steps_gbs32": 4365,
+  "packing_efficiency_percent": 99.704,
+  "packs": 139671,
+  "sampled_assistant_loss_tokens": 3054640,
+  "sampled_packs_checked_for_loss_masks": 128,
+  "samples": 742654,
+  "source_samples": 742665
+}

packed-65k-qwen3-30b-a3b/dataset_info.json

@@ -0,0 +1,32 @@
+{
+  "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+  "tokenizer": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/qwen3-30b-a3b-thinking-generation-spans",
+  "max_len": 65536,
+  "pack_size": 65536,
+  "shards": 8,
+  "kept_samples": 742658,
+  "train_samples": 742658,
+  "val_samples": 0,
+  "train_packs": 132145,
+  "val_packs": 0,
+  "train_tokens": 8637920650,
+  "val_tokens": 0,
+  "train_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/train",
+  "val_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/val.parquet",
+  "packed_metadata": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k-qwen3-30b-a3b/packed_metadata.json",
+  "turn_mask": {
+    "rows_seen": 742665,
+    "rows_kept": 742658,
+    "rows_over_max_len": 2,
+    "rows_no_trainable": 0,
+    "rows_with_masked_turns": 33291,
+    "rows_fully_masked": 5,
+    "masked_turns": 38145,
+    "masked_tokens": 53754611,
+    "trainable_tokens": 3343677840,
+    "turn_flags_on_non_assistant": 0,
+    "span_turn_count_mismatch": 0,
+    "span_bad_start": 0,
+    "span_multiple_headers": 0
+  }
+}

packed-65k-qwen3-30b-a3b/packed_metadata.json

@@ -0,0 +1,66 @@
+[
+  {
+    "dataset_max_seqlen": 64906,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.59,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 29831
+  },
+  {
+    "dataset_max_seqlen": 61011,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.63,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 17157
+  },
+  {
+    "dataset_max_seqlen": 59708,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.6,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 15580
+  },
+  {
+    "dataset_max_seqlen": 64963,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.61,
+    "packing_efficiency": 99.74,
+    "pack_size": 65536,
+    "min_packed_seqlen": 62753
+  },
+  {
+    "dataset_max_seqlen": 60128,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.62,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 23191
+  },
+  {
+    "dataset_max_seqlen": 59611,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.61,
+    "packing_efficiency": 99.74,
+    "pack_size": 65536,
+    "min_packed_seqlen": 58734
+  },
+  {
+    "dataset_max_seqlen": 62547,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.65,
+    "packing_efficiency": 99.74,
+    "pack_size": 65536,
+    "min_packed_seqlen": 57978
+  },
+  {
+    "dataset_max_seqlen": 54299,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.64,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 36718
+  }
+]

packed-65k-qwen3-30b-a3b/train/train_000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07042c26a52406c961462816bc82b1908cb3ec76fa3670649639bbdf305329c3
+size 1176334827

packed-65k-qwen3-30b-a3b/train/train_001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e26a1363f4ba6bde304524254b5fb00f93dadcbc692238d8948cc78464ed05d
+size 1176783641

packed-65k-qwen3-30b-a3b/train/train_002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc6e7f0f83559a8c57ef5ee8f616ea0820a1149bc39fca0ad263a4155cf7b1c5
+size 1178061667

packed-65k-qwen3-30b-a3b/train/train_003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7c89c95f76b341c2d2a4c635b5d4925abe799ae34dc9c0616289fa3747bfe5b
+size 1185828561

packed-65k-qwen3-30b-a3b/train/train_004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e186bcb69dea510a3dbff9e1be6abe107f7a53730a90b63428f037778ba302a5
+size 1175906994

packed-65k-qwen3-30b-a3b/train/train_005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b640ced4c6d809b77fe5ba79926693c55adf2fae36f3462d2b082bad6c6ef02a
+size 1177347517

packed-65k-qwen3-30b-a3b/train/train_006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4526c41636b8a1bb7dddd823d6aef0f61c8a94cdfd06b7f80bd3acdebd31352d
+size 1170255094

packed-65k-qwen3-30b-a3b/train/train_007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4707d841c54acaa45c6c0e8e385a59feedc382a1590bc660e929c0c2b3f0801e
+size 1177487403

packed-65k-qwen3-30b-a3b/validation-summary.json

@@ -0,0 +1,12 @@
+{
+  "dropped_overlength_samples": 7,
+  "effective_tokens": 8637920650,
+  "one_epoch_steps_gbs16": 8260,
+  "one_epoch_steps_gbs32": 4130,
+  "packing_efficiency_percent": 99.7421,
+  "packs": 132145,
+  "sampled_assistant_loss_tokens": 3269634,
+  "sampled_packs_checked_for_loss_masks": 128,
+  "samples": 742658,
+  "source_samples": 742665
+}

packed-65k/dataset_info.json

@@ -0,0 +1,32 @@
+{
+  "src": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/raw/v0_9_final.jsonl",
+  "tokenizer": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/env/qwen35-qwen36-thinking-generation-spans",
+  "max_len": 65536,
+  "pack_size": 65536,
+  "shards": 8,
+  "kept_samples": 742656,
+  "train_samples": 742656,
+  "val_samples": 0,
+  "train_packs": 137384,
+  "val_packs": 0,
+  "train_tokens": 8978945130,
+  "val_tokens": 0,
+  "train_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/train",
+  "val_parquet": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/val.parquet",
+  "packed_metadata": "/NHNHOME/WORKSPACE/wbl-workspace/agentic-rl-260803/datasets/v0.9/packed-65k/packed_metadata.json",
+  "turn_mask": {
+    "rows_seen": 742665,
+    "rows_kept": 742656,
+    "rows_over_max_len": 4,
+    "rows_no_trainable": 0,
+    "rows_with_masked_turns": 33291,
+    "rows_fully_masked": 5,
+    "masked_turns": 38145,
+    "masked_tokens": 54577131,
+    "trainable_tokens": 3408327232,
+    "turn_flags_on_non_assistant": 0,
+    "span_turn_count_mismatch": 0,
+    "span_bad_start": 0,
+    "span_multiple_headers": 0
+  }
+}

packed-65k/packed_metadata.json

@@ -0,0 +1,66 @@
+[
+  {
+    "dataset_max_seqlen": 58722,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.38,
+    "packing_efficiency": 99.71,
+    "pack_size": 65536,
+    "min_packed_seqlen": 30980
+  },
+  {
+    "dataset_max_seqlen": 63648,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.42,
+    "packing_efficiency": 99.71,
+    "pack_size": 65536,
+    "min_packed_seqlen": 42264
+  },
+  {
+    "dataset_max_seqlen": 61151,
+    "max_samples_per_bin": 12,
+    "packing_factor": 5.39,
+    "packing_efficiency": 99.71,
+    "pack_size": 65536,
+    "min_packed_seqlen": 59514
+  },
+  {
+    "dataset_max_seqlen": 60748,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.39,
+    "packing_efficiency": 99.72,
+    "pack_size": 65536,
+    "min_packed_seqlen": 28348
+  },
+  {
+    "dataset_max_seqlen": 61299,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.41,
+    "packing_efficiency": 99.72,
+    "pack_size": 65536,
+    "min_packed_seqlen": 44958
+  },
+  {
+    "dataset_max_seqlen": 61168,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.39,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 23337
+  },
+  {
+    "dataset_max_seqlen": 63891,
+    "max_samples_per_bin": 13,
+    "packing_factor": 5.44,
+    "packing_efficiency": 99.73,
+    "pack_size": 65536,
+    "min_packed_seqlen": 47238
+  },
+  {
+    "dataset_max_seqlen": 56757,
+    "max_samples_per_bin": 14,
+    "packing_factor": 5.43,
+    "packing_efficiency": 99.71,
+    "pack_size": 65536,
+    "min_packed_seqlen": 46447
+  }
+]

packed-65k/train/train_000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c9f3838cce60bfa3d0e590de587bc2175c5018ac877e851df90164b849caa32
+size 1199525185

packed-65k/train/train_001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9840f3c8badf9efb0d76e6262067edf4a002a99891e0fdfece6a8a498b0fda5
+size 1213487156