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": 74958.03101589713,
-      "required_pages_per_second_upper_estimate": 0.006497172149065229
+      "required_token_positions_per_second": 75173.78962979905,
+      "required_pages_per_second_upper_estimate": 0.006515873558883041
     },
     {
       "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": 9369.75387698714,
-      "required_pages_per_second_upper_estimate": 0.003737734911834666
+      "required_token_positions_per_second": 9396.723703724881,
+      "required_pages_per_second_upper_estimate": 0.0037484935789551944
     },
     {
       "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": 18739.50775397428,
-      "required_pages_per_second_upper_estimate": 0.004335055314809688
+      "required_token_positions_per_second": 18793.447407449763,
+      "required_pages_per_second_upper_estimate": 0.004347533304335762
     },
     {
       "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": 28109.26163096142,
-      "required_pages_per_second_upper_estimate": 0.0017117065554274654
+      "required_token_positions_per_second": 28190.171111174644,
+      "required_pages_per_second_upper_estimate": 0.001716633518273215
     },
     {
       "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": 28109.26163096142,
-      "required_pages_per_second_upper_estimate": 0.0021639287717677568
+      "required_token_positions_per_second": 28190.171111174644,
+      "required_pages_per_second_upper_estimate": 0.002170157407526347
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28109.26163096142,
-      "required_pages_per_second_upper_estimate": 0.349022953809571
+      "required_token_positions_per_second": 28190.171111174644,
+      "required_pages_per_second_upper_estimate": 0.3500275787672082
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12015560151573368,
-    "producer_p95_seconds": 0.3225278470017656,
-    "gpu_update_seconds": 0.9835904038669818,
-    "producer_batches_per_second": 8.322541665850308,
-    "gpu_batches_per_second": 1.0166833633883616,
-    "producer_headroom": 8.185972118313488,
+    "producer_batch_seconds": 0.08918089049225841,
+    "producer_p95_seconds": 0.2568429990024015,
+    "gpu_update_seconds": 0.980767370689718,
+    "producer_batches_per_second": 11.213164552184054,
+    "gpu_batches_per_second": 1.01960977687987,
+    "producer_headroom": 10.997505914956704,
     "sustainable": true,
-    "buffer_seconds": 60.982605039752876,
+    "buffer_seconds": 60.80757698276251,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324167.44212716713,
-    "upload_bytes_per_second": 5579801.162934172,
+    "generation_bytes_per_second": 310157.67965546664,
+    "upload_bytes_per_second": 5499046.131477833,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.55878684865246,
+    "estimated_full_shards_per_hour": 66.55261795280455,
     "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": 1.0363285590010491,
-      "bytes": 11683428,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.0476808559997153,
       "bytes": 995690,
@@ -562,6 +556,12 @@
       "bytes": 2386369,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2013670170017576,
+      "bytes": 6446691,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 20600, "loss": 0.665947962552309, "lr": 0.0003, "grad_norm": 0.12847631823769204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3014375678, "tokens_per_second": 148735.51007187043, "data_wait_s": 1.3164000847609714e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20700, "loss": 0.6963246762752533, "lr": 0.0003, "grad_norm": 0.14299341454153447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3029005928, "tokens_per_second": 148941.26953540367, "data_wait_s": 1.072499799192883e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20800, "loss": 0.7289977855980396, "lr": 0.0003, "grad_norm": 0.15577767153490452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3043656574, "tokens_per_second": 148297.71208878103, "data_wait_s": 1.1731997801689431e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 20900, "loss": 0.6284867636859417, "lr": 0.0003, "grad_norm": 0.1758226288225565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3058158501, "tokens_per_second": 147687.78176371922, "data_wait_s": 8.006001735338941e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59db163a52ceccf0b1773645567b9893e75659cabe6cc36b5b9974be46bb1e0e
-size 438067
+oid sha256:b5f76a958f99b1be8a7de491c6dacf742283e974c60ad269466510d9e7b9e560
+size 440185