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": 76644.32262976514,
-      "required_pages_per_second_upper_estimate": 0.005302432078857132
+      "required_token_positions_per_second": 76541.73475827192,
+      "required_pages_per_second_upper_estimate": 0.005295334811870574
     },
     {
       "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": 9580.540328720643,
-      "required_pages_per_second_upper_estimate": 0.004011172171203185
+      "required_token_positions_per_second": 9567.71684478399,
+      "required_pages_per_second_upper_estimate": 0.0040058032462637035
     },
     {
       "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": 19161.080657441285,
-      "required_pages_per_second_upper_estimate": 0.003990747789267344
+      "required_token_positions_per_second": 19135.43368956798,
+      "required_pages_per_second_upper_estimate": 0.003985406202215361
     },
     {
       "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": 28741.620986161928,
-      "required_pages_per_second_upper_estimate": 0.0017230162922062385
+      "required_token_positions_per_second": 28703.150534351975,
+      "required_pages_per_second_upper_estimate": 0.0017207100473612105
     },
     {
       "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": 28741.620986161928,
-      "required_pages_per_second_upper_estimate": 0.002234534610643958
+      "required_token_positions_per_second": 28703.150534351975,
+      "required_pages_per_second_upper_estimate": 0.0022315437022293687
     },
     {
       "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": 28741.620986161928,
-      "required_pages_per_second_upper_estimate": 0.36866665366224044
+      "required_token_positions_per_second": 28703.150534351975,
+      "required_pages_per_second_upper_estimate": 0.3681731960127753
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07485346717186303,
-    "producer_p95_seconds": 0.1944179880010779,
-    "gpu_update_seconds": 0.9619499197109144,
-    "producer_batches_per_second": 13.35943460980915,
-    "gpu_batches_per_second": 1.0395551571962505,
-    "producer_headroom": 12.851107050289123,
+    "producer_batch_seconds": 0.502953613492167,
+    "producer_p95_seconds": 0.2389487710006506,
+    "gpu_update_seconds": 0.9632392084245537,
+    "producer_batches_per_second": 1.9882549268444094,
+    "gpu_batches_per_second": 1.0381637201371519,
+    "producer_headroom": 1.915165101879828,
     "sustainable": true,
-    "buffer_seconds": 60.6028449417876,
+    "buffer_seconds": 60.68407013074688,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415418.98322456214,
-    "upload_bytes_per_second": 5350857.921276327,
+    "generation_bytes_per_second": 352967.8199659083,
+    "upload_bytes_per_second": 5385605.177399345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.13924334099434,
+    "estimated_full_shards_per_hour": 75.73867749436319,
     "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": 0.9467493070001183,
-      "bytes": 2673204,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3979449759999625,
       "bytes": 6163668,
@@ -562,6 +556,12 @@
       "bytes": 897417,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.0840730929994606,
+      "bytes": 6199927,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -116,3 +116,4 @@
 {"step": 568700, "loss": 0.5584443807601929, "lr": 0.0003, "grad_norm": 0.16764500432431012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83240882348, "tokens_per_second": 151947.46953504998, "data_wait_s": 1.3409000530373305e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568800, "loss": 0.5851945281028748, "lr": 0.0003, "grad_norm": 0.16060189837159555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83255376272, "tokens_per_second": 151518.0398185791, "data_wait_s": 1.0312000085832551e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568900, "loss": 0.5981356501579285, "lr": 0.0003, "grad_norm": 0.16078664710792331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83270053882, "tokens_per_second": 152990.8233905173, "data_wait_s": 8.864000847097486e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 569000, "loss": 0.5846968293190002, "lr": 0.0003, "grad_norm": 0.16389304236523466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83284694450, "tokens_per_second": 150266.70682100102, "data_wait_s": 1.3680000847671181e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f873f47bb79a8048fc63d817a9588c56801e64480aa0b27808a61751ec3e4469
-size 251655
+oid sha256:e464786c76857abd5c2480791e487fb2d4e1843f69d953feeb9c1c388332251d
+size 253773