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": 77158.38677200767,
-      "required_pages_per_second_upper_estimate": 0.0053379962290106145
+      "required_token_positions_per_second": 75768.24415210997,
+      "required_pages_per_second_upper_estimate": 0.005241822936990816
     },
     {
       "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": 9644.798346500958,
-      "required_pages_per_second_upper_estimate": 0.004038075661387803
+      "required_token_positions_per_second": 9471.030519013746,
+      "required_pages_per_second_upper_estimate": 0.003965322700703777
     },
     {
       "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": 19289.596693001917,
-      "required_pages_per_second_upper_estimate": 0.004017514290278853
+      "required_token_positions_per_second": 18942.061038027492,
+      "required_pages_per_second_upper_estimate": 0.00394513177847923
     },
     {
       "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": 28934.395039502873,
-      "required_pages_per_second_upper_estimate": 0.001734572802355092
+      "required_token_positions_per_second": 28413.09155704124,
+      "required_pages_per_second_upper_estimate": 0.001703321454565846
     },
     {
       "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": 28934.395039502873,
-      "required_pages_per_second_upper_estimate": 0.002249521945367767
+      "required_token_positions_per_second": 28413.09155704124,
+      "required_pages_per_second_upper_estimate": 0.002208992892578066
     },
     {
       "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": 28934.395039502873,
-      "required_pages_per_second_upper_estimate": 0.3711393522338461
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28413.09155704124,
+      "required_pages_per_second_upper_estimate": 0.36445263089289825
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08717399907820322,
-    "producer_p95_seconds": 0.23552601399933337,
-    "gpu_update_seconds": 0.9555409733728106,
-    "producer_batches_per_second": 11.471310374357227,
-    "gpu_batches_per_second": 1.046527598361649,
-    "producer_headroom": 10.961307080974926,
+    "producer_batch_seconds": 0.39745253627336297,
+    "producer_p95_seconds": 0.25326214500091737,
+    "gpu_update_seconds": 0.9730725691885634,
+    "producer_batches_per_second": 2.5160236977635293,
+    "gpu_batches_per_second": 1.0276725823582624,
+    "producer_headroom": 2.448273643722067,
     "sustainable": true,
-    "buffer_seconds": 60.19908132248707,
+    "buffer_seconds": 60.330499289690934,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312412.06207660784,
+    "generation_bytes_per_second": 280295.04552975786,
     "upload_bytes_per_second": 1879252.6533477928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.03635594104458,
+    "estimated_full_shards_per_hour": 60.144791835971375,
     "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

@@ -208,3 +208,4 @@
 {"step": 665800, "loss": 0.5446346402168274, "lr": 0.0001, "grad_norm": 0.17459977320421943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97456844897, "tokens_per_second": 145374.92154866926, "data_wait_s": 1.3748998753726482e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 665900, "loss": 0.5673105716705322, "lr": 0.0001, "grad_norm": 0.18219381795551592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97471508796, "tokens_per_second": 152615.54873536606, "data_wait_s": 1.2598997273016721e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666000, "loss": 0.5200461745262146, "lr": 0.0001, "grad_norm": 0.1869376745769584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97486206069, "tokens_per_second": 153150.03655282865, "data_wait_s": 1.5675999748054892e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 666100, "loss": 0.579171895980835, "lr": 0.0001, "grad_norm": 0.1892237160171339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97500725009, "tokens_per_second": 150352.89336690045, "data_wait_s": 1.7468002624809742e-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 sha256:75060fc1c8bb43cefdd6325283cb7bfeba18c3af1e0d7c4eb3279f7ce5519fc1
-size 448077
+oid sha256:2dd40f8d928c9f61c9b7741d5122a938db208ad4cab8494190cd46fc4962e5f6
+size 450195