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": 77502.5041545277,
-      "required_pages_per_second_upper_estimate": 0.005361803067995685
+      "required_token_positions_per_second": 75552.18958640698,
+      "required_pages_per_second_upper_estimate": 0.005226875780819823
     },
     {
       "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": 9687.813019315963,
-      "required_pages_per_second_upper_estimate": 0.004056085006646934
+      "required_token_positions_per_second": 9444.023698300873,
+      "required_pages_per_second_upper_estimate": 0.003954015508837844
     },
     {
       "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": 19375.626038631926,
-      "required_pages_per_second_upper_estimate": 0.004035431934227173
+      "required_token_positions_per_second": 18888.047396601745,
+      "required_pages_per_second_upper_estimate": 0.003933882161405761
     },
     {
       "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": 29063.43905794789,
-      "required_pages_per_second_upper_estimate": 0.0017423087942220637
+      "required_token_positions_per_second": 28332.071094902618,
+      "required_pages_per_second_upper_estimate": 0.0016984644015717182
     },
     {
       "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": 29063.43905794789,
-      "required_pages_per_second_upper_estimate": 0.0022595545501972163
+      "required_token_positions_per_second": 28332.071094902618,
+      "required_pages_per_second_upper_estimate": 0.0022026939080181385
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29063.43905794789,
-      "required_pages_per_second_upper_estimate": 0.3727945903457869
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28332.071094902618,
+      "required_pages_per_second_upper_estimate": 0.3634133873975785
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11852404745306444,
-    "producer_p95_seconds": 0.261483563001093,
-    "gpu_update_seconds": 0.9512982942202494,
-    "producer_batches_per_second": 8.437106405735936,
-    "gpu_batches_per_second": 1.051194989075083,
-    "producer_headroom": 8.026204931931336,
+    "producer_batch_seconds": 0.45325259092214765,
+    "producer_p95_seconds": 0.26878805000160355,
+    "gpu_update_seconds": 0.9758552386582958,
+    "producer_batches_per_second": 2.2062753087974376,
+    "gpu_batches_per_second": 1.0247421547635496,
+    "producer_headroom": 2.1530053180124287,
     "sustainable": true,
-    "buffer_seconds": 60.88309083009596,
+    "buffer_seconds": 60.503024796814344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316136.1575087695,
+    "generation_bytes_per_second": 312701.1176121372,
     "upload_bytes_per_second": 1867524.5728969113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.83546012828172,
+    "estimated_full_shards_per_hour": 67.0983805301007,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 666300, "loss": 0.5639391541481018, "lr": 0.0001, "grad_norm": 0.20244867513736692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97529618966, "tokens_per_second": 150769.23351798995, "data_wait_s": 1.2085998605471104e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666400, "loss": 0.615656316280365, "lr": 0.0001, "grad_norm": 0.19269532786653254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97544316740, "tokens_per_second": 153093.79737789245, "data_wait_s": 1.0648000170476735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666500, "loss": 0.5665801763534546, "lr": 0.0001, "grad_norm": 0.2004456924753406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97558969927, "tokens_per_second": 151801.83431139376, "data_wait_s": 1.2847998732468113e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 666600, "loss": 0.5367074012756348, "lr": 0.0001, "grad_norm": 0.19993975317060272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97573544513, "tokens_per_second": 151572.8148915651, "data_wait_s": 1.4500998076982796e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b27191624937b09b44151796fdaba0da1a6bc208d903074a60c8cb0234119d98
-size 458667
+oid sha256:40b6a63f9ca527f2591648dfe3db1c156bbf2e9a40b09d0c6c2277f6d2bb5e9c
+size 460785