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": 77471.78373052676,
-      "required_pages_per_second_upper_estimate": 0.005359677757782095
+      "required_token_positions_per_second": 76443.3441000466,
+      "required_pages_per_second_upper_estimate": 0.005288527917836765
     },
     {
       "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": 9683.972966315845,
-      "required_pages_per_second_upper_estimate": 0.00405447725664521
+      "required_token_positions_per_second": 9555.418012505825,
+      "required_pages_per_second_upper_estimate": 0.004000653982017659
     },
     {
       "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": 19367.94593263169,
-      "required_pages_per_second_upper_estimate": 0.0040338323706853385
+      "required_token_positions_per_second": 19110.83602501165,
+      "required_pages_per_second_upper_estimate": 0.003980283157372313
     },
     {
       "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": 29051.918898947537,
-      "required_pages_per_second_upper_estimate": 0.0017416181782802568
+      "required_token_positions_per_second": 28666.254037517476,
+      "required_pages_per_second_upper_estimate": 0.001718498158713671
     },
     {
       "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": 29051.918898947537,
-      "required_pages_per_second_upper_estimate": 0.0022586589085067635
+      "required_token_positions_per_second": 28666.254037517476,
+      "required_pages_per_second_upper_estimate": 0.0022286751618910247
     },
     {
       "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": 29051.918898947537,
-      "required_pages_per_second_upper_estimate": 0.37264682211551337
+      "required_token_positions_per_second": 28666.254037517476,
+      "required_pages_per_second_upper_estimate": 0.3676999273677541
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.11722742142217157,
-    "producer_p95_seconds": 0.3198310450025019,
-    "gpu_update_seconds": 0.9516755191341802,
-    "producer_batches_per_second": 8.53042733405093,
-    "gpu_batches_per_second": 1.0507783166575353,
-    "producer_headroom": 8.11819886156932,
+    "producer_batch_seconds": 0.35405055258593165,
+    "producer_p95_seconds": 0.30035731100360863,
+    "gpu_update_seconds": 0.9644789990284459,
+    "producer_batches_per_second": 2.824455413771145,
+    "gpu_batches_per_second": 1.0368292114264133,
+    "producer_headroom": 2.7241279302744688,
     "sustainable": true,
-    "buffer_seconds": 60.907233224587536,
+    "buffer_seconds": 60.762176938792095,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393362.914305699,
-    "upload_bytes_per_second": 4861289.193374157,
+    "generation_bytes_per_second": 329818.73177540675,
+    "upload_bytes_per_second": 4956906.683021564,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.406524390013,
+    "estimated_full_shards_per_hour": 70.77142205187465,
     "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": [
-    {
-      "seconds": 1.1109868420026032,
-      "bytes": 2592499,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.24596934100191,
       "bytes": 5407802,
@@ -562,6 +556,12 @@
       "bytes": 7548903,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2989959650003584,
+      "bytes": 11854087,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 591300, "loss": 0.6259649991989136, "lr": 0.0003, "grad_norm": 0.1636940153074632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86548471983, "tokens_per_second": 152535.70387104695, "data_wait_s": 1.1028001608792692e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 591400, "loss": 0.5447350740432739, "lr": 0.0003, "grad_norm": 0.16897899380884174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86563185633, "tokens_per_second": 152892.044087327, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 591500, "loss": 0.6079108119010925, "lr": 0.0003, "grad_norm": 0.17477587074679687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86577718105, "tokens_per_second": 150134.66179609686, "data_wait_s": 1.1860000086016953e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 591600, "loss": 0.577265739440918, "lr": 0.0003, "grad_norm": 0.19341440023267018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86592303768, "tokens_per_second": 151194.54178578706, "data_wait_s": 1.3630000466946512e-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:f7825d55d23fdb1e34f44f7bb33b6e730d755e646a0003439de72c05bf8fadf4
-size 734238
+oid sha256:8d51ddb0982054123288320e6251fcb41dc06807ca2997856547af37d4ee1c02
+size 736356