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": 77490.03042101624,
-      "required_pages_per_second_upper_estimate": 0.005360940106168313
+      "required_token_positions_per_second": 76919.55409073165,
+      "required_pages_per_second_upper_estimate": 0.00532147322995177
     },
     {
       "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": 9686.25380262703,
-      "required_pages_per_second_upper_estimate": 0.004055432195179425
+      "required_token_positions_per_second": 9614.944261341456,
+      "required_pages_per_second_upper_estimate": 0.0040255763793557095
     },
     {
       "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": 19372.50760525406,
-      "required_pages_per_second_upper_estimate": 0.004034782446793182
+      "required_token_positions_per_second": 19229.88852268291,
+      "required_pages_per_second_upper_estimate": 0.004005078653011743
     },
     {
       "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": 29058.76140788109,
-      "required_pages_per_second_upper_estimate": 0.001742028376242917
+      "required_token_positions_per_second": 28844.832784024366,
+      "required_pages_per_second_upper_estimate": 0.0017292036818928015
     },
     {
       "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": 29058.76140788109,
-      "required_pages_per_second_upper_estimate": 0.0022591908834793313
+      "required_token_positions_per_second": 28844.832784024366,
+      "required_pages_per_second_upper_estimate": 0.002242558873947042
     },
     {
       "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": 29058.76140788109,
-      "required_pages_per_second_upper_estimate": 0.3727345904731993
+      "required_token_positions_per_second": 28844.832784024366,
+      "required_pages_per_second_upper_estimate": 0.3699905437850254
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11979438224994965,
-    "producer_p95_seconds": 0.31162487999972655,
-    "gpu_update_seconds": 0.9514514267115847,
-    "producer_batches_per_second": 8.347636852565516,
-    "gpu_batches_per_second": 1.05102580323644,
-    "producer_headroom": 7.942370993043662,
+    "producer_batch_seconds": 0.17390677089866813,
+    "producer_p95_seconds": 0.31997386000148254,
+    "gpu_update_seconds": 0.9585078966140782,
+    "producer_batches_per_second": 5.750207394642956,
+    "gpu_batches_per_second": 1.0432882228018072,
+    "producer_headroom": 5.511619194933939,
     "sustainable": true,
-    "buffer_seconds": 60.89289130954142,
+    "buffer_seconds": 60.385997486686925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391604.23308132886,
+    "generation_bytes_per_second": 396400.62473933166,
     "upload_bytes_per_second": 5423799.182112922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.02915233926677,
+    "estimated_full_shards_per_hour": 85.05834633479083,
     "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

@@ -339,3 +339,4 @@
 {"step": 678900, "loss": 0.6020879149436951, "lr": 0.0001, "grad_norm": 0.2556691093215879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99373500312, "tokens_per_second": 152638.41731959954, "data_wait_s": 1.2704003893304616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679000, "loss": 0.5635417103767395, "lr": 0.0001, "grad_norm": 0.1881923509732763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99388288592, "tokens_per_second": 153790.0230992682, "data_wait_s": 1.0928997653536499e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679100, "loss": 0.5356208086013794, "lr": 0.0001, "grad_norm": 0.20511929759650416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99402851670, "tokens_per_second": 151488.33874070636, "data_wait_s": 1.654700463404879e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 679200, "loss": 0.6018109321594238, "lr": 0.0001, "grad_norm": 0.21745714831770144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99417364559, "tokens_per_second": 150898.36260275112, "data_wait_s": 1.4440003724303097e-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 sha256:11757a6c10e23050f3e9998def4bc8ddbd187d36d60679fd47651aac7bf0034e
-size 727101
+oid sha256:2771e30cf2f1199af842dfab648797d34ac7343ec1a560569968e7b5f46bf5b0
+size 729219