Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75046.23646551954,
-      "required_pages_per_second_upper_estimate": 0.0065048175738838315
+      "required_token_positions_per_second": 74286.13081836392,
+      "required_pages_per_second_upper_estimate": 0.006438933542858532
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9380.779558189943,
-      "required_pages_per_second_upper_estimate": 0.0037421332209150882
+      "required_token_positions_per_second": 9285.76635229549,
+      "required_pages_per_second_upper_estimate": 0.0037042310325097693
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18761.559116379885,
-      "required_pages_per_second_upper_estimate": 0.004340156509411501
+      "required_token_positions_per_second": 18571.53270459098,
+      "required_pages_per_second_upper_estimate": 0.004296197243394762
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28142.33867456983,
-      "required_pages_per_second_upper_estimate": 0.0017137207738413818
+      "required_token_positions_per_second": 27857.299056886466,
+      "required_pages_per_second_upper_estimate": 0.0016963633566117614
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28142.33867456983,
-      "required_pages_per_second_upper_estimate": 0.0021664751341478493
+      "required_token_positions_per_second": 27857.299056886466,
+      "required_pages_per_second_upper_estimate": 0.0021445319953384214
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28142.33867456983,
-      "required_pages_per_second_upper_estimate": 0.34943365998944376
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27857.299056886466,
+      "required_pages_per_second_upper_estimate": 0.3458944219040499
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.0904514655858577,
-    "producer_p95_seconds": 0.27085126099700574,
-    "gpu_update_seconds": 0.9824343427784654,
-    "producer_batches_per_second": 11.055652813616238,
-    "gpu_batches_per_second": 1.0178797263660964,
-    "producer_headroom": 10.86145300593196,
+    "producer_batch_seconds": 0.30812177457031,
+    "producer_p95_seconds": 0.2872173069990822,
+    "gpu_update_seconds": 0.9924867426501376,
+    "producer_batches_per_second": 3.2454700788172017,
+    "gpu_batches_per_second": 1.0075701337126182,
+    "producer_headroom": 3.22108602689377,
     "sustainable": true,
-    "buffer_seconds": 60.91092925226486,
+    "buffer_seconds": 60.541691301658396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416622.90701490693,
+    "generation_bytes_per_second": 378719.90408817725,
     "upload_bytes_per_second": 5836369.178753451,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.39757736049086,
+    "estimated_full_shards_per_hour": 81.26447526916493,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 374900, "loss": 0.5981489419937134, "lr": 0.0003, "grad_norm": 0.14830377690576205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54874543805, "tokens_per_second": 150633.8736510711, "data_wait_s": 1.1036001524189487e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375000, "loss": 0.5523220244795084, "lr": 0.0003, "grad_norm": 0.1553186573423817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54889201011, "tokens_per_second": 149205.15722583298, "data_wait_s": 1.4123997971182689e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375100, "loss": 0.6094298958778381, "lr": 0.0003, "grad_norm": 0.16543400737552919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54903847811, "tokens_per_second": 125171.26088617968, "data_wait_s": 1.2852997315349057e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 375200, "loss": 0.5757131744176149, "lr": 0.0003, "grad_norm": 0.15580247337755893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54918289514, "tokens_per_second": 147168.59554686316, "data_wait_s": 1.3649998436449096e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9386e3d24e9270351239c1f2d28d0dc55c39f28a8b5a146b4927d88bd5c884e
-size 429015
+oid sha256:90778f617232f7734c21544a474dbc44b5c9e53dcb24c7fff3f991cba92726f9
+size 431133