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": 77489.43882378846,
-      "required_pages_per_second_upper_estimate": 0.0053608991780993995
+      "required_token_positions_per_second": 77191.26322178313,
+      "required_pages_per_second_upper_estimate": 0.005340270698089954
     },
     {
       "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": 9686.179852973557,
-      "required_pages_per_second_upper_estimate": 0.00405540123400376
+      "required_token_positions_per_second": 9648.90790272289,
+      "required_pages_per_second_upper_estimate": 0.004039796246760634
     },
     {
       "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": 19372.359705947114,
-      "required_pages_per_second_upper_estimate": 0.004034751643267912
+      "required_token_positions_per_second": 19297.81580544578,
+      "required_pages_per_second_upper_estimate": 0.0040192261146483384
     },
     {
       "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": 29058.53955892067,
-      "required_pages_per_second_upper_estimate": 0.0017420150767364837
+      "required_token_positions_per_second": 28946.723708168676,
+      "required_pages_per_second_upper_estimate": 0.00173531188721682
     },
     {
       "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": 29058.53955892067,
-      "required_pages_per_second_upper_estimate": 0.002259173635698469
+      "required_token_positions_per_second": 28946.723708168676,
+      "required_pages_per_second_upper_estimate": 0.002250480445128451
     },
     {
       "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": 29058.53955892067,
-      "required_pages_per_second_upper_estimate": 0.3727317448329379
+      "required_token_positions_per_second": 28946.723708168676,
+      "required_pages_per_second_upper_estimate": 0.3712974911579979
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06761386993744622,
-    "producer_p95_seconds": 0.2236622919990623,
-    "gpu_update_seconds": 0.9514586906179306,
-    "producer_batches_per_second": 14.789864874250828,
-    "gpu_batches_per_second": 1.0510177791854987,
-    "producer_headroom": 14.071945467670819,
+    "producer_batch_seconds": 0.4318409104685941,
+    "producer_p95_seconds": 0.23562351999862585,
+    "gpu_update_seconds": 0.9551339999213044,
+    "producer_batches_per_second": 2.31566758905471,
+    "gpu_batches_per_second": 1.0469735137503138,
+    "producer_headroom": 2.2117728468219484,
     "sustainable": true,
-    "buffer_seconds": 60.89335619954756,
+    "buffer_seconds": 60.17344199504218,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356750.43968752085,
-    "upload_bytes_per_second": 1939755.3517491233,
+    "generation_bytes_per_second": 336644.7679893691,
+    "upload_bytes_per_second": 1893391.7791501223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.55033963174075,
+    "estimated_full_shards_per_hour": 72.2361305214005,
     "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": 1.0431637549991137,
-      "bytes": 11688824,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.3166948790003516,
       "bytes": 11002922,
@@ -562,6 +556,12 @@
       "bytes": 11733880,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.0325014549998741,
+      "bytes": 941600,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 664100, "loss": 0.5199372172355652, "lr": 0.0001, "grad_norm": 0.17990135708849617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97207173670, "tokens_per_second": 150201.06743699362, "data_wait_s": 1.4229000953491777e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 664200, "loss": 0.5283790230751038, "lr": 0.0001, "grad_norm": 0.19205047311701712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97221728887, "tokens_per_second": 151822.64087532778, "data_wait_s": 1.0331001249141991e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 664300, "loss": 0.5534589290618896, "lr": 0.0001, "grad_norm": 0.1947474065961574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97236417590, "tokens_per_second": 141860.8271409519, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 664400, "loss": 0.5364418625831604, "lr": 0.0001, "grad_norm": 0.1788650947893196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97251111143, "tokens_per_second": 153055.23395418428, "data_wait_s": 9.680003131506965e-06, "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 sha256:3d561c5c37d9f343f5bcf26d6f76f66831fef721fedacf194354dc6b155536a8
-size 411288
+oid sha256:025c079288a2d60e1f81098f0200c13e0f8b08c9587859ce2579c9ab0ffbfbab
+size 413406