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": 77436.00385182936,
-      "required_pages_per_second_upper_estimate": 0.005357202422753052
+      "required_token_positions_per_second": 76849.2751745828,
+      "required_pages_per_second_upper_estimate": 0.00531661117146304
     },
     {
       "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": 9679.50048147867,
-      "required_pages_per_second_upper_estimate": 0.004052604720639988
+      "required_token_positions_per_second": 9606.15939682285,
+      "required_pages_per_second_upper_estimate": 0.004021898340030601
     },
     {
       "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": 19359.00096295734,
-      "required_pages_per_second_upper_estimate": 0.004031969369396885
+      "required_token_positions_per_second": 19212.3187936457,
+      "required_pages_per_second_upper_estimate": 0.004001419341798205
     },
     {
       "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": 29038.50144443601,
-      "required_pages_per_second_upper_estimate": 0.0017408138223695576
+      "required_token_positions_per_second": 28818.47819046855,
+      "required_pages_per_second_upper_estimate": 0.0017276237642502646
     },
     {
       "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": 29038.50144443601,
-      "required_pages_per_second_upper_estimate": 0.0022576157604356373
+      "required_token_positions_per_second": 28818.47819046855,
+      "required_pages_per_second_upper_estimate": 0.0022405099202196833
     },
     {
       "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": 29038.50144443601,
-      "required_pages_per_second_upper_estimate": 0.3724747174155797
+      "required_token_positions_per_second": 28818.47819046855,
+      "required_pages_per_second_upper_estimate": 0.3696524953562493
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11354683278108268,
-    "producer_p95_seconds": 0.3281363150017569,
-    "gpu_update_seconds": 0.9521152478513164,
-    "producer_batches_per_second": 8.80693873626569,
-    "gpu_batches_per_second": 1.0502930209937793,
-    "producer_headroom": 8.385220657690967,
+    "producer_batch_seconds": 0.234811860975924,
+    "producer_p95_seconds": 0.37940049699682277,
+    "gpu_update_seconds": 0.9593844552535853,
+    "producer_batches_per_second": 4.258728651286202,
+    "gpu_batches_per_second": 1.042335003995535,
+    "producer_headroom": 4.085758067187049,
     "sustainable": true,
-    "buffer_seconds": 60.93537586248425,
+    "buffer_seconds": 60.441220680975874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 657735.2158284521,
-    "upload_bytes_per_second": 3779961.4946783804,
+    "generation_bytes_per_second": 620553.1916752339,
+    "upload_bytes_per_second": 3781593.1290231417,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.56733301229559,
+    "estimated_full_shards_per_hour": 66.57813459726697,
     "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": 0.9759389660030138,
-      "bytes": 6031085,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.012582071001816,
       "bytes": 5087118,
@@ -562,6 +556,12 @@
       "bytes": 6356475,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.0511947449995205,
+      "bytes": 6504509,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -406,3 +406,4 @@
 {"step": 525600, "loss": 0.5734371542930603, "lr": 0.0003, "grad_norm": 0.1713987125041783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76930290681, "tokens_per_second": 152874.47791848576, "data_wait_s": 1.586099824635312e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525700, "loss": 0.5619049072265625, "lr": 0.0003, "grad_norm": 0.15671392922051514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76944945407, "tokens_per_second": 153603.49633318934, "data_wait_s": 7.855996955186129e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 525800, "loss": 0.5962910056114197, "lr": 0.0003, "grad_norm": 0.17840752645790417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76959784739, "tokens_per_second": 153971.1352422353, "data_wait_s": 1.5576995792798698e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 525900, "loss": 0.5798600316047668, "lr": 0.0003, "grad_norm": 0.16361460049218451, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76974555081, "tokens_per_second": 154452.6054116465, "data_wait_s": 1.4794000890105963e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39f509e1f14f77480c742ebaa1b042fc3dacf072782ee8ec0f96f1c18a455442
-size 870573
+oid sha256:2c20bf55746788541697366a68c774bb08b7d889748dc8c50c4bfb98cace4900
+size 872691