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": 76270.64855498832,
-      "required_pages_per_second_upper_estimate": 0.005276580439320762
+      "required_token_positions_per_second": 77280.73115412582,
+      "required_pages_per_second_upper_estimate": 0.0053464602972436255
     },
     {
       "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": 9533.83106937354,
-      "required_pages_per_second_upper_estimate": 0.0039916159797148045
+      "required_token_positions_per_second": 9660.091394265728,
+      "required_pages_per_second_upper_estimate": 0.00404447854113176
     },
     {
       "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": 19067.66213874708,
-      "required_pages_per_second_upper_estimate": 0.00397129117543535
+      "required_token_positions_per_second": 19320.182788531456,
+      "required_pages_per_second_upper_estimate": 0.0040238845673680745
     },
     {
       "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": 28601.49320812062,
-      "required_pages_per_second_upper_estimate": 0.0017146158458753904
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.0017373231869421907
     },
     {
       "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": 28601.49320812062,
-      "required_pages_per_second_upper_estimate": 0.0022236402922582036
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.0022530888469578757
     },
     {
       "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": 28601.49320812062,
-      "required_pages_per_second_upper_estimate": 0.36686924498301227
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.3717278406228395
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09654542103127284,
-    "producer_p95_seconds": 0.1625932280003326,
-    "gpu_update_seconds": 0.9666628171759263,
-    "producer_batches_per_second": 10.357819038109344,
-    "gpu_batches_per_second": 1.0344868781872332,
-    "producer_headroom": 10.012518531177221,
+    "producer_batch_seconds": 0.23549360834378064,
+    "producer_p95_seconds": 0.1584173479996025,
+    "gpu_update_seconds": 0.9540282409202315,
+    "producer_batches_per_second": 4.2463997517086325,
+    "gpu_batches_per_second": 1.0481870002458473,
+    "producer_headroom": 4.051185285366694,
     "sustainable": true,
-    "buffer_seconds": 60.89975748208336,
+    "buffer_seconds": 60.10377917797459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 515602.8343726077,
-    "upload_bytes_per_second": 3710302.153971347,
+    "generation_bytes_per_second": 486911.6116432644,
+    "upload_bytes_per_second": 3768783.117588168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.63636563666985,
+    "estimated_full_shards_per_hour": 104.47989713643503,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 1048802,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 0.9894983370004411,
+      "bytes": 7420139,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -45,3 +45,4 @@
 {"step": 444700, "loss": 0.5960000157356262, "lr": 0.0003, "grad_norm": 0.1548911055979621, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65090874113, "tokens_per_second": 152302.72494620606, "data_wait_s": 1.0871999620576389e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444800, "loss": 0.5636985301971436, "lr": 0.0003, "grad_norm": 0.16352203214652344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65105497414, "tokens_per_second": 151367.44674651648, "data_wait_s": 1.705900012893835e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444900, "loss": 0.5851013660430908, "lr": 0.0003, "grad_norm": 0.16064967921121015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65119922459, "tokens_per_second": 149715.8220490589, "data_wait_s": 7.629000720044132e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 445000, "loss": 0.6301952004432678, "lr": 0.0003, "grad_norm": 0.20991914998299785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65134552525, "tokens_per_second": 152765.13751353952, "data_wait_s": 1.3879000107408501e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fca7ef2d0dc5b910267acf01e95af40e17d7924357d6dddb01e2dc76fc42475
-size 99711
+oid sha256:2b268d56036879e6f92a6466d419696a3b8269deaec78c7b62c5914ddf2b7afc
+size 101829