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": 75568.14143254272,
-      "required_pages_per_second_upper_estimate": 0.005227979366548876
+      "required_token_positions_per_second": 77146.65142505743,
+      "required_pages_per_second_upper_estimate": 0.00533718435047884
     },
     {
       "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": 9446.01767906784,
-      "required_pages_per_second_upper_estimate": 0.00395485034694592
+      "required_token_positions_per_second": 9643.331428132178,
+      "required_pages_per_second_upper_estimate": 0.004037461493299534
     },
     {
       "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": 18892.03535813568,
-      "required_pages_per_second_upper_estimate": 0.003934712748623661
+      "required_token_positions_per_second": 19286.662856264356,
+      "required_pages_per_second_upper_estimate": 0.004016903249456897
     },
     {
       "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": 28338.053037203517,
-      "required_pages_per_second_upper_estimate": 0.00169882300987876
+      "required_token_positions_per_second": 28929.994284396533,
+      "required_pages_per_second_upper_estimate": 0.001734308983806031
     },
     {
       "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": 28338.053037203517,
-      "required_pages_per_second_upper_estimate": 0.00220315897772025
+      "required_token_positions_per_second": 28929.994284396533,
+      "required_pages_per_second_upper_estimate": 0.0022491798060151253
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28338.053037203517,
-      "required_pages_per_second_upper_estimate": 0.36349011733050524
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28929.994284396533,
+      "required_pages_per_second_upper_estimate": 0.3710829040725046
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.47558302648451445,
-    "producer_p95_seconds": 0.2851251060019422,
-    "gpu_update_seconds": 0.9756492432173239,
-    "producer_batches_per_second": 2.1026822748321132,
-    "gpu_batches_per_second": 1.024958515523854,
-    "producer_headroom": 2.0514803701664324,
+    "producer_batch_seconds": 0.11588557424238388,
+    "producer_p95_seconds": 0.28689809199931915,
+    "gpu_update_seconds": 0.9556863277678564,
+    "producer_batches_per_second": 8.629201749550125,
+    "gpu_batches_per_second": 1.0463684275317033,
+    "producer_headroom": 8.246810131595522,
     "sustainable": true,
-    "buffer_seconds": 60.49025307947409,
+    "buffer_seconds": 60.208238649374955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317483.0981236007,
+    "generation_bytes_per_second": 317444.9159052798,
     "upload_bytes_per_second": 6143690.370607286,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.12448222905174,
+    "estimated_full_shards_per_hour": 68.11628921383662,
     "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

@@ -253,3 +253,4 @@
 {"step": 510300, "loss": 0.5971457362174988, "lr": 0.0003, "grad_norm": 0.16460713640988167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74690558967, "tokens_per_second": 152727.17945273404, "data_wait_s": 1.7277998267672956e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510400, "loss": 0.5947121381759644, "lr": 0.0003, "grad_norm": 0.14776955372029305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74705240141, "tokens_per_second": 150967.06811949794, "data_wait_s": 1.730199801386334e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510500, "loss": 0.6076563596725464, "lr": 0.0003, "grad_norm": 0.15903708875644673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74719951999, "tokens_per_second": 153463.67566932138, "data_wait_s": 1.971500023500994e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 510600, "loss": 0.5998474955558777, "lr": 0.0003, "grad_norm": 0.16425945716733695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74734713757, "tokens_per_second": 149271.93040092214, "data_wait_s": 1.3688000763067976e-05, "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 sha256:2999023f451b6e9e76d102d0471c1d7456d161ecb168054d459a6af3ab58460f
-size 544170
+oid sha256:a8f8063b9e3d81a216cabdc8c19f929bb253e574a875991f8d1d9e6035c7f2d5
+size 546288