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": 75986.42926582096,
-      "required_pages_per_second_upper_estimate": 0.005256917489416555
+      "required_token_positions_per_second": 76118.41913935372,
+      "required_pages_per_second_upper_estimate": 0.005266048855126252
     },
     {
       "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": 9498.30365822762,
-      "required_pages_per_second_upper_estimate": 0.003976741394564716
+      "required_token_positions_per_second": 9514.802392419215,
+      "required_pages_per_second_upper_estimate": 0.003983649070037989
     },
     {
       "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": 18996.60731645524,
-      "required_pages_per_second_upper_estimate": 0.003956492329793634
+      "required_token_positions_per_second": 19029.60478483843,
+      "required_pages_per_second_upper_estimate": 0.003963364832256093
     },
     {
       "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": 28494.910974682858,
-      "required_pages_per_second_upper_estimate": 0.001708226403722444
+      "required_token_positions_per_second": 28544.407177257643,
+      "required_pages_per_second_upper_estimate": 0.001711193625490477
     },
     {
       "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": 28494.910974682858,
-      "required_pages_per_second_upper_estimate": 0.0022153539924141172
+      "required_token_positions_per_second": 28544.407177257643,
+      "required_pages_per_second_upper_estimate": 0.00221920210445352
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28494.910974682858,
-      "required_pages_per_second_upper_estimate": 0.36550212253155884
+      "required_token_positions_per_second": 28544.407177257643,
+      "required_pages_per_second_upper_estimate": 0.36613700667330645
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11583690349243625,
-    "producer_p95_seconds": 0.3631641710016993,
-    "gpu_update_seconds": 0.9702785183138378,
-    "producer_batches_per_second": 8.632827448338142,
-    "gpu_batches_per_second": 1.0306319073597678,
-    "producer_headroom": 8.376247025432562,
+    "producer_batch_seconds": 0.31791389319579366,
+    "producer_p95_seconds": 0.3595181920027244,
+    "gpu_update_seconds": 0.9685960485467064,
+    "producer_batches_per_second": 3.145505815891254,
+    "gpu_batches_per_second": 1.0324221345940987,
+    "producer_headroom": 3.046724503952953,
     "sustainable": true,
-    "buffer_seconds": 60.15726813545795,
+    "buffer_seconds": 60.0529550098958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468189.5408586986,
+    "generation_bytes_per_second": 536395.7813486883,
     "upload_bytes_per_second": 5263471.663121072,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.46257657356948,
+    "estimated_full_shards_per_hour": 115.09804802270399,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -358,3 +358,4 @@
 {"step": 432900, "loss": 0.5553624033927917, "lr": 0.0003, "grad_norm": 0.15352128821337596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63364505407, "tokens_per_second": 151727.28399990578, "data_wait_s": 1.2948003131896257e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433000, "loss": 0.6017171144485474, "lr": 0.0003, "grad_norm": 0.15152019661519708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63379170427, "tokens_per_second": 151419.2022181481, "data_wait_s": 1.349899685010314e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433100, "loss": 0.5756955146789551, "lr": 0.0003, "grad_norm": 0.17255215265487533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63393783835, "tokens_per_second": 151430.9350176385, "data_wait_s": 1.2305004929658026e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 433200, "loss": 0.5947546362876892, "lr": 0.0003, "grad_norm": 0.15544372888172547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63408219132, "tokens_per_second": 149199.0484397043, "data_wait_s": 1.79109993041493e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e048239e4a417f2d3a9688614aa599ca0b99afeb64f2dd673af51a49473d939f
-size 768126
+oid sha256:ae62d603c7c6f19766e5036b6e912a28f45cfaefe51e05b30eec9ef08bed3c66
+size 770244