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": 77228.98378152134,
-      "required_pages_per_second_upper_estimate": 0.005342880293936401
+      "required_token_positions_per_second": 77467.37180956296,
+      "required_pages_per_second_upper_estimate": 0.005359372530852753
     },
     {
       "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": 9653.622972690167,
-      "required_pages_per_second_upper_estimate": 0.004041770348094075
+      "required_token_positions_per_second": 9683.42147619537,
+      "required_pages_per_second_upper_estimate": 0.00405424635924819
     },
     {
       "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": 19307.245945380335,
-      "required_pages_per_second_upper_estimate": 0.004021190164107192
+      "required_token_positions_per_second": 19366.84295239074,
+      "required_pages_per_second_upper_estimate": 0.004033602648988694
     },
     {
       "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": 28960.868918070504,
-      "required_pages_per_second_upper_estimate": 0.0017361598709519496
+      "required_token_positions_per_second": 29050.26442858611,
+      "required_pages_per_second_upper_estimate": 0.0017415189953083191
     },
     {
       "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": 28960.868918070504,
-      "required_pages_per_second_upper_estimate": 0.002251580172980115
+      "required_token_positions_per_second": 29050.26442858611,
+      "required_pages_per_second_upper_estimate": 0.0022585302807134086
     },
     {
       "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": 28960.868918070504,
-      "required_pages_per_second_upper_estimate": 0.3714789307226755
+      "required_token_positions_per_second": 29050.26442858611,
+      "required_pages_per_second_upper_estimate": 0.37262560034614883
     }
   ],
   "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": 0.08572673410915854,
-    "producer_p95_seconds": 0.24008956699981354,
-    "gpu_update_seconds": 0.9546674886798261,
-    "producer_batches_per_second": 11.66497254784801,
-    "gpu_batches_per_second": 1.0474851315853047,
-    "producer_headroom": 11.136170047773172,
+    "producer_batch_seconds": 0.08939321218750251,
+    "producer_p95_seconds": 0.2577405970005202,
+    "gpu_update_seconds": 0.9517297189485735,
+    "producer_batches_per_second": 11.186531678741975,
+    "gpu_batches_per_second": 1.0507184761496713,
+    "producer_headroom": 10.646554650618413,
     "sustainable": true,
-    "buffer_seconds": 60.144051786829046,
+    "buffer_seconds": 60.910702012708704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281145.2695846663,
-    "upload_bytes_per_second": 5084360.319831612,
+    "generation_bytes_per_second": 267036.0321375891,
+    "upload_bytes_per_second": 5115542.3874711245,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.32723012595169,
+    "estimated_full_shards_per_hour": 57.29971621604686,
     "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": 1.1321277869992628,
-      "bytes": 2736110,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.411381372001415,
       "bytes": 15991084,
@@ -562,6 +556,12 @@
       "bytes": 2330980,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2888210940000135,
+      "bytes": 6181272,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 903200, "loss": 0.5517054796218872, "lr": 3e-05, "grad_norm": 0.343599887765593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132202381888, "tokens_per_second": 152948.24850813881, "data_wait_s": 1.4413999451790005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 903300, "loss": 0.5157208442687988, "lr": 3e-05, "grad_norm": 0.3094748408921095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132216944601, "tokens_per_second": 151477.93788325097, "data_wait_s": 2.0908999431412667e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 903400, "loss": 0.5840005874633789, "lr": 3e-05, "grad_norm": 0.355907021136958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132231624024, "tokens_per_second": 152967.58851264016, "data_wait_s": 1.4622000890085474e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 903500, "loss": 0.5231959223747253, "lr": 3e-05, "grad_norm": 0.33438638105747265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132246414721, "tokens_per_second": 144868.6990460812, "data_wait_s": 1.669799894443713e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10a58d0d1d9da2413c566f3449441a0219f2e03f0a26fbd515e7e8dbc9b280d4
-size 392226
+oid sha256:62639336ba0200de0dfcac660a28744240a5e22b7b1418766004fc1b4367b37d
+size 394344