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": 77568.14314135184,
-      "required_pages_per_second_upper_estimate": 0.005366344125407612
+      "required_token_positions_per_second": 76619.61853602182,
+      "required_pages_per_second_upper_estimate": 0.005300722992330574
     },
     {
       "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": 9696.01789266898,
-      "required_pages_per_second_upper_estimate": 0.004059520215782603
+      "required_token_positions_per_second": 9577.452317002728,
+      "required_pages_per_second_upper_estimate": 0.004009879285181911
     },
     {
       "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": 19392.03578533796,
-      "required_pages_per_second_upper_estimate": 0.004038849651711918
+      "required_token_positions_per_second": 19154.904634005456,
+      "required_pages_per_second_upper_estimate": 0.003989461486458352
     },
     {
       "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": 29088.05367800694,
-      "required_pages_per_second_upper_estimate": 0.0017437844031102552
+      "required_token_positions_per_second": 28732.356951008183,
+      "required_pages_per_second_upper_estimate": 0.0017224609274441342
     },
     {
       "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": 29088.05367800694,
-      "required_pages_per_second_upper_estimate": 0.002261468228638536
+      "required_token_positions_per_second": 28732.356951008183,
+      "required_pages_per_second_upper_estimate": 0.0022338143726589386
     },
     {
       "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": 29088.05367800694,
-      "required_pages_per_second_upper_estimate": 0.3731103202627845
+      "required_token_positions_per_second": 28732.356951008183,
+      "required_pages_per_second_upper_estimate": 0.3685478245662342
     }
   ],
   "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": 0.0800855250625574,
-    "producer_p95_seconds": 0.16931907000071078,
-    "gpu_update_seconds": 0.9504932954969157,
-    "producer_batches_per_second": 12.486650979922622,
-    "gpu_batches_per_second": 1.0520852748121725,
-    "producer_headroom": 11.868478039626446,
+    "producer_batch_seconds": 0.43279563547655897,
+    "producer_p95_seconds": 0.18170442000064213,
+    "gpu_update_seconds": 0.9622600765799642,
+    "producer_batches_per_second": 2.310559344940903,
+    "gpu_batches_per_second": 1.0392200864803307,
+    "producer_headroom": 2.2233590122053855,
     "sustainable": true,
-    "buffer_seconds": 60.83157091180261,
+    "buffer_seconds": 60.62238482453774,
     "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": 360279.5100767769,
-    "upload_bytes_per_second": 5547302.971057902,
+    "generation_bytes_per_second": 319782.58540255186,
+    "upload_bytes_per_second": 5602472.206029002,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.30759598472099,
+    "estimated_full_shards_per_hour": 68.61789866979043,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 2411720,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.251686485999926,
+      "bytes": 10374348,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 563400, "loss": 0.5847949385643005, "lr": 0.0003, "grad_norm": 0.16201373228750962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82464199918, "tokens_per_second": 151849.51189046344, "data_wait_s": 1.1978000657109078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 563500, "loss": 0.539670467376709, "lr": 0.0003, "grad_norm": 0.15464426605453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82478879592, "tokens_per_second": 153389.1666795691, "data_wait_s": 7.966000339365564e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 563600, "loss": 0.5937413573265076, "lr": 0.0003, "grad_norm": 0.16478871309476983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82493543819, "tokens_per_second": 153064.2229277863, "data_wait_s": 1.3582000065071043e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 563700, "loss": 0.5554092526435852, "lr": 0.0003, "grad_norm": 0.16144070464643215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82508202407, "tokens_per_second": 152566.43981824303, "data_wait_s": 1.6103999769256916e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c27ffb7698a1bb6b28bb8ea725f18924c957ded80dd3c7e7a0c490f7a38b20b
-size 138618
+oid sha256:64059c6c7c928c7cb0dbf4753ba975204eea595f5cbdf0d39d8c5613a34aa58b
+size 140736