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": 76307.95135286232,
-      "required_pages_per_second_upper_estimate": 0.005279161133432101
+      "required_token_positions_per_second": 76986.59462324777,
+      "required_pages_per_second_upper_estimate": 0.005326111249546702
     },
     {
       "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": 9538.49391910779,
-      "required_pages_per_second_upper_estimate": 0.00399356821752716
+      "required_token_positions_per_second": 9623.324327905972,
+      "required_pages_per_second_upper_estimate": 0.004029084938230583
     },
     {
       "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": 19076.98783821558,
-      "required_pages_per_second_upper_estimate": 0.003973233472699405
+      "required_token_positions_per_second": 19246.648655811943,
+      "required_pages_per_second_upper_estimate": 0.004008569346748087
     },
     {
       "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": 28615.481757323367,
-      "required_pages_per_second_upper_estimate": 0.001715454438040817
+      "required_token_positions_per_second": 28869.972983717915,
+      "required_pages_per_second_upper_estimate": 0.0017307107984775674
     },
     {
       "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": 28615.481757323367,
-      "required_pages_per_second_upper_estimate": 0.0022247278404295915
+      "required_token_positions_per_second": 28869.972983717915,
+      "required_pages_per_second_upper_estimate": 0.0022445134138931633
     },
     {
       "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": 28615.481757323367,
-      "required_pages_per_second_upper_estimate": 0.3670486750724512
+      "required_token_positions_per_second": 28869.972983717915,
+      "required_pages_per_second_upper_estimate": 0.3703130152732509
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0954549049924367,
-    "producer_p95_seconds": 0.32441666300292127,
-    "gpu_update_seconds": 0.9661902684173483,
-    "producer_batches_per_second": 10.47615101685172,
-    "gpu_batches_per_second": 1.0349928297643,
-    "producer_headroom": 10.12195516295264,
+    "producer_batch_seconds": 0.1515634279452911,
+    "producer_p95_seconds": 0.32563139100238914,
+    "gpu_update_seconds": 0.9576732203938298,
+    "producer_batches_per_second": 6.597897748531814,
+    "gpu_batches_per_second": 1.044197518218964,
+    "producer_headroom": 6.318629984665662,
     "sustainable": true,
-    "buffer_seconds": 60.869986910292944,
+    "buffer_seconds": 60.33341288481128,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244150.69501919032,
-    "upload_bytes_per_second": 3666832.9400221235,
+    "generation_bytes_per_second": 244346.71693659545,
+    "upload_bytes_per_second": 3683783.0727058197,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.38905561382086,
+    "estimated_full_shards_per_hour": 52.4311173541393,
     "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": 1.0129257990010956,
-      "bytes": 2288801,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.3661311769974418,
       "bytes": 11305021,
@@ -562,6 +556,12 @@
       "bytes": 10458494,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 1.8176103080040775,
+      "bytes": 7189151,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 880900, "loss": 0.5186980366706848, "lr": 3e-05, "grad_norm": 0.32294184104601764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128939072513, "tokens_per_second": 149594.84237675718, "data_wait_s": 1.2589000107254833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881000, "loss": 0.5027999877929688, "lr": 3e-05, "grad_norm": 0.29615730230163156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128953639377, "tokens_per_second": 150685.19613681379, "data_wait_s": 1.4551995263900608e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881100, "loss": 0.5287010073661804, "lr": 3e-05, "grad_norm": 0.3909493589346569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128968269669, "tokens_per_second": 151252.52617220164, "data_wait_s": 1.4672994439024478e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 881200, "loss": 0.5220397114753723, "lr": 3e-05, "grad_norm": 0.3119779935049428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128982934902, "tokens_per_second": 152550.5468513958, "data_wait_s": 1.4767996617592871e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c0fb2ab7d0dec679be0d26dd1624b58439fffbb44a09763cd4952b432e36abc
-size 823977
+oid sha256:2de7f39017fedd1b21dc4adce282c16d6d454a867f85a91ca5d3326d7d80d48b
+size 826095