Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77457.07488158996,
-      "required_pages_per_second_upper_estimate": 0.0053586601654834155
+      "required_token_positions_per_second": 77390.10854074381,
+      "required_pages_per_second_upper_estimate": 0.005354027278640361
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9682.134360198745,
-      "required_pages_per_second_upper_estimate": 0.004053707470658442
+      "required_token_positions_per_second": 9673.763567592976,
+      "required_pages_per_second_upper_estimate": 0.004050202794596434
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19364.26872039749,
-      "required_pages_per_second_upper_estimate": 0.004033066504351563
+      "required_token_positions_per_second": 19347.527135185952,
+      "required_pages_per_second_upper_estimate": 0.0040295796736572916
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29046.403080596232,
-      "required_pages_per_second_upper_estimate": 0.001741287513392263
+      "required_token_positions_per_second": 29021.29070277893,
+      "required_pages_per_second_upper_estimate": 0.0017397820646864964
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29046.403080596232,
-      "required_pages_per_second_upper_estimate": 0.00225823007789147
+      "required_token_positions_per_second": 29021.29070277893,
+      "required_pages_per_second_upper_estimate": 0.0022562777009738516
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29046.403080596232,
-      "required_pages_per_second_upper_estimate": 0.37257607112012714
+      "required_token_positions_per_second": 29021.29070277893,
+      "required_pages_per_second_upper_estimate": 0.3722539565010573
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08488724995305574,
-    "producer_p95_seconds": 0.25194475000171224,
-    "gpu_update_seconds": 0.9518562392487625,
-    "producer_batches_per_second": 11.780332152979616,
-    "gpu_batches_per_second": 1.0505788151257318,
-    "producer_headroom": 11.213182660236455,
+    "producer_batch_seconds": 0.06694714627352027,
+    "producer_p95_seconds": 0.2514671460012323,
+    "gpu_update_seconds": 0.9526798888153024,
+    "producer_batches_per_second": 14.93715648333067,
+    "gpu_batches_per_second": 1.0496705259975019,
+    "producer_headroom": 14.230328577756236,
     "sustainable": true,
-    "buffer_seconds": 60.9187993119208,
+    "buffer_seconds": 60.01883299536405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358293.4313251985,
+    "generation_bytes_per_second": 400506.1695543668,
     "upload_bytes_per_second": 5526136.612045022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.88142971817938,
+    "estimated_full_shards_per_hour": 85.93930067990544,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 501800, "loss": 0.5342846512794495, "lr": 0.0003, "grad_norm": 0.15386128369866137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73446603551, "tokens_per_second": 152406.28316549293, "data_wait_s": 1.8790000467561185e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501900, "loss": 0.5525649785995483, "lr": 0.0003, "grad_norm": 0.1397344511084202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73461282059, "tokens_per_second": 153024.28076948592, "data_wait_s": 1.2282998795853928e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502000, "loss": 0.562541663646698, "lr": 0.0003, "grad_norm": 0.16451517640575014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73475920503, "tokens_per_second": 152698.6983736639, "data_wait_s": 1.0954998288070783e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 502100, "loss": 0.5456511974334717, "lr": 0.0003, "grad_norm": 0.1837708465292662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73490554006, "tokens_per_second": 152031.23970780458, "data_wait_s": 9.715000487631187e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90a4321b23015a1616e5e77e42abe5caeb0eb2510a1c652ec301ee030a186145
-size 362574
+oid sha256:39336eb904ce2e810376354f623b810ff1fdcfa5154083a45f42d298b05ece3f
+size 364692