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": 77399.79067672219,
-      "required_pages_per_second_upper_estimate": 0.0053546971112729954
+      "required_token_positions_per_second": 76499.11329940458,
+      "required_pages_per_second_upper_estimate": 0.005292386160450721
     },
     {
       "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": 9674.973834590273,
-      "required_pages_per_second_upper_estimate": 0.004050709508114953
+      "required_token_positions_per_second": 9562.389162425572,
+      "required_pages_per_second_upper_estimate": 0.004003572656914893
     },
     {
       "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": 19349.947669180547,
-      "required_pages_per_second_upper_estimate": 0.004030083807054592
+      "required_token_positions_per_second": 19124.778324851144,
+      "required_pages_per_second_upper_estimate": 0.003983186970745708
     },
     {
       "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": 29024.92150377082,
-      "required_pages_per_second_upper_estimate": 0.0017399997256620481
+      "required_token_positions_per_second": 28687.167487276718,
+      "required_pages_per_second_upper_estimate": 0.0017197518880937488
     },
     {
       "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": 29024.92150377082,
-      "required_pages_per_second_upper_estimate": 0.002256559979780765
+      "required_token_positions_per_second": 28687.167487276718,
+      "required_pages_per_second_upper_estimate": 0.002230301090621262
     },
     {
       "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": 29024.92150377082,
-      "required_pages_per_second_upper_estimate": 0.37230052851773093
+      "required_token_positions_per_second": 28687.167487276718,
+      "required_pages_per_second_upper_estimate": 0.3679681826461528
     }
   ],
   "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": 0.08285795357022607,
-    "producer_p95_seconds": 0.312756253999396,
-    "gpu_update_seconds": 0.9525607156735313,
-    "producer_batches_per_second": 12.068847429986938,
-    "gpu_batches_per_second": 1.0498018483713405,
-    "producer_headroom": 11.496309945263016,
+    "producer_batch_seconds": 0.4906027926641059,
+    "producer_p95_seconds": 0.30866892600170104,
+    "gpu_update_seconds": 0.9637758768711617,
+    "producer_batches_per_second": 2.03830882121508,
+    "gpu_batches_per_second": 1.0375856296034693,
+    "producer_headroom": 1.9644728715007878,
     "sustainable": true,
-    "buffer_seconds": 60.01132508743247,
+    "buffer_seconds": 60.71788024288319,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258905.64254173508,
-    "upload_bytes_per_second": 5470537.114867328,
+    "generation_bytes_per_second": 245816.8824328196,
+    "upload_bytes_per_second": 5548142.56339119,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.555123874559776,
+    "estimated_full_shards_per_hour": 52.74658064592782,
     "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": 1.102126538000448,
-      "bytes": 5859413,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2403170809993753,
       "bytes": 7520295,
@@ -562,6 +556,12 @@
       "bytes": 1235313,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.3809676959972421,
+      "bytes": 13424902,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -317,3 +317,4 @@
 {"step": 471900, "loss": 0.5733816027641296, "lr": 0.0003, "grad_norm": 0.15933333398687086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69071504283, "tokens_per_second": 131852.69168520567, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472000, "loss": 0.5416136980056763, "lr": 0.0003, "grad_norm": 0.15735503096387818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69086078993, "tokens_per_second": 151613.48434305555, "data_wait_s": 1.518900171504356e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472100, "loss": 0.5716539025306702, "lr": 0.0003, "grad_norm": 0.18243784115190018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69100692261, "tokens_per_second": 144324.6328924901, "data_wait_s": 5.646450822998304, "gpu_peak_gb": 89.37899398803711}
+{"step": 472200, "loss": 0.5773496627807617, "lr": 0.0003, "grad_norm": 0.1731953844509603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69115198521, "tokens_per_second": 150481.56125245686, "data_wait_s": 1.2366999726509675e-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 sha256:0f997096869403102a720c3a291fdcd16092ccd7ec75016654669c2956f981db
-size 680505
+oid sha256:1c236805b1a5fddeb12d3a1cc48a8831daa9e57c1c12e63896062a8e301af8de
+size 682623