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": 77474.20817935525,
-      "required_pages_per_second_upper_estimate": 0.0053598454867775465
+      "required_token_positions_per_second": 76929.33644384316,
+      "required_pages_per_second_upper_estimate": 0.0053221499958382105
     },
     {
       "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": 9684.276022419406,
-      "required_pages_per_second_upper_estimate": 0.00405460413990724
+      "required_token_positions_per_second": 9616.167055480395,
+      "required_pages_per_second_upper_estimate": 0.004026088337726838
     },
     {
       "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": 19368.552044838812,
-      "required_pages_per_second_upper_estimate": 0.004033958607873828
+      "required_token_positions_per_second": 19232.33411096079,
+      "required_pages_per_second_upper_estimate": 0.004005588004555525
     },
     {
       "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": 29052.82806725822,
-      "required_pages_per_second_upper_estimate": 0.0017416726815322626
+      "required_token_positions_per_second": 28848.501166441183,
+      "required_pages_per_second_upper_estimate": 0.001729423595817394
     },
     {
       "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": 29052.82806725822,
-      "required_pages_per_second_upper_estimate": 0.00225872959234391
+      "required_token_positions_per_second": 28848.501166441183,
+      "required_pages_per_second_upper_estimate": 0.0022428440745444404
     },
     {
       "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": 29052.82806725822,
-      "required_pages_per_second_upper_estimate": 0.3726584839504139
+      "required_token_positions_per_second": 28848.501166441183,
+      "required_pages_per_second_upper_estimate": 0.37003759785586615
     }
   ],
   "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": 0.5190950927891436,
-    "producer_p95_seconds": 0.3838786889973562,
-    "gpu_update_seconds": 0.9516457377572333,
-    "producer_batches_per_second": 1.926429307252573,
-    "gpu_batches_per_second": 1.050811200349328,
-    "producer_headroom": 1.8332782393375306,
+    "producer_batch_seconds": 0.1581329905156963,
+    "producer_p95_seconds": 0.322475362001569,
+    "gpu_update_seconds": 0.9583860125170836,
+    "producer_batches_per_second": 6.32379111239751,
+    "gpu_batches_per_second": 1.043420904457508,
+    "producer_headroom": 6.060632948201622,
     "sustainable": true,
-    "buffer_seconds": 60.90532721646293,
+    "buffer_seconds": 60.378318788576266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277744.25531368994,
+    "generation_bytes_per_second": 275165.3709405064,
     "upload_bytes_per_second": 5190993.263093734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.5974516349604,
+    "estimated_full_shards_per_hour": 59.04408308183092,
     "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

@@ -387,3 +387,4 @@
 {"step": 523700, "loss": 0.5777722597122192, "lr": 0.0003, "grad_norm": 0.15813718426402518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76651725396, "tokens_per_second": 152906.96139242983, "data_wait_s": 8.773000445216894e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 523800, "loss": 0.5599973201751709, "lr": 0.0003, "grad_norm": 0.16208737839385862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76666386923, "tokens_per_second": 133169.49670360115, "data_wait_s": 1.5987003280315548e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523900, "loss": 0.6106423735618591, "lr": 0.0003, "grad_norm": 0.19980792504345476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76681058358, "tokens_per_second": 153079.29437360534, "data_wait_s": 1.4077995729167014e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 524000, "loss": 0.5886540412902832, "lr": 0.0003, "grad_norm": 0.16838592874414218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76695771128, "tokens_per_second": 153939.2906792162, "data_wait_s": 1.3355995179153979e-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 sha256:1aed230e9e5ec4c19a390fa8ea5df0f67d403f04b39d3083ca42b54c496288e7
-size 829548
+oid sha256:8546f7aa5f15707fe6a5ebc4fdbfe1cd8e7e89b777944e8a02761a8032e19857
+size 831666