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": 76940.0009421512,
-      "required_pages_per_second_upper_estimate": 0.005322887790576205
+      "required_token_positions_per_second": 77430.1918528613,
+      "required_pages_per_second_upper_estimate": 0.005356800335178747
     },
     {
       "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": 9617.5001177689,
-      "required_pages_per_second_upper_estimate": 0.004026646463069529
+      "required_token_positions_per_second": 9678.773981607663,
+      "required_pages_per_second_upper_estimate": 0.00405230055031504
     },
     {
       "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": 19235.0002355378,
-      "required_pages_per_second_upper_estimate": 0.004006143287994483
+      "required_token_positions_per_second": 19357.547963215326,
+      "required_pages_per_second_upper_estimate": 0.004031666747868804
     },
     {
       "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": 28852.500353306696,
-      "required_pages_per_second_upper_estimate": 0.0017296633409635768
+      "required_token_positions_per_second": 29036.321944822987,
+      "required_pages_per_second_upper_estimate": 0.00174068316469566
     },
     {
       "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": 28852.500353306696,
-      "required_pages_per_second_upper_estimate": 0.00224315499373267
+      "required_token_positions_per_second": 29036.321944822987,
+      "required_pages_per_second_upper_estimate": 0.0022574463139273305
     },
     {
       "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": 28852.500353306696,
-      "required_pages_per_second_upper_estimate": 0.37008889513098464
+      "required_token_positions_per_second": 29036.321944822987,
+      "required_pages_per_second_upper_estimate": 0.3724467611347082
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.28520037475800564,
-    "producer_p95_seconds": 0.23843101999955252,
-    "gpu_update_seconds": 0.9582531725653839,
-    "producer_batches_per_second": 3.506306753097735,
-    "gpu_batches_per_second": 1.0435655509731878,
-    "producer_headroom": 3.359929570143335,
+    "producer_batch_seconds": 0.08475807853912443,
+    "producer_p95_seconds": 0.2233907280024141,
+    "gpu_update_seconds": 0.9521867147133448,
+    "producer_batches_per_second": 11.798285393390541,
+    "gpu_batches_per_second": 1.050214190712637,
+    "producer_headroom": 11.234170607982982,
     "sustainable": true,
-    "buffer_seconds": 60.369949871619184,
+    "buffer_seconds": 60.93994974165407,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356381.37129932694,
-    "upload_bytes_per_second": 4300937.506376537,
+    "generation_bytes_per_second": 356228.72999961517,
+    "upload_bytes_per_second": 4288190.16309335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.47114614710671,
+    "estimated_full_shards_per_hour": 76.43839287749616,
     "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": 1.1455854699997872,
-      "bytes": 11238485,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 0.9324971470014134,
       "bytes": 1088704,
@@ -562,6 +556,12 @@
       "bytes": 13329088,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1372946299998148,
+      "bytes": 9925526,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -176,3 +176,4 @@
 {"step": 702600, "loss": 0.5840139389038086, "lr": 8e-05, "grad_norm": 0.22603571300496714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102841166411, "tokens_per_second": 152453.53514426816, "data_wait_s": 1.4353998267324641e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702700, "loss": 0.5287814736366272, "lr": 8e-05, "grad_norm": 0.22470357132468263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102855798182, "tokens_per_second": 152124.93571930652, "data_wait_s": 1.7035999917425215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702800, "loss": 0.5411754250526428, "lr": 8e-05, "grad_norm": 0.21744803486100425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102870373097, "tokens_per_second": 151645.50629856766, "data_wait_s": 1.3174001651350409e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 702900, "loss": 0.5611929893493652, "lr": 8e-05, "grad_norm": 0.22294607376503367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102884966095, "tokens_per_second": 152282.90303154627, "data_wait_s": 1.485399843659252e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2325b2de09a550e10c420b876336d36184760c93197642e3b0942792f683851
-size 379518
+oid sha256:c5a733df0af79aca87e0d351cab012fb69a00a396e6d810ba080d7832e504676
+size 381636