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": 77047.90253709067,
-      "required_pages_per_second_upper_estimate": 0.005330352673280313
+      "required_token_positions_per_second": 76638.2072425708,
+      "required_pages_per_second_upper_estimate": 0.005302009002181374
     },
     {
       "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": 9630.987817136334,
-      "required_pages_per_second_upper_estimate": 0.0040322934811394835
+      "required_token_positions_per_second": 9579.77590532135,
+      "required_pages_per_second_upper_estimate": 0.004010852123088876
     },
     {
       "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": 19261.97563427267,
-      "required_pages_per_second_upper_estimate": 0.004011761552161853
+      "required_token_positions_per_second": 19159.5518106427,
+      "required_pages_per_second_upper_estimate": 0.0039904293707976
     },
     {
       "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": 28892.963451409003,
-      "required_pages_per_second_upper_estimate": 0.0017320890419112367
+      "required_token_positions_per_second": 28739.32771596405,
+      "required_pages_per_second_upper_estimate": 0.0017228788141594965
     },
     {
       "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": 28892.963451409003,
-      "required_pages_per_second_upper_estimate": 0.0022463008216317653
+      "required_token_positions_per_second": 28739.32771596405,
+      "required_pages_per_second_upper_estimate": 0.0022343563189731025
     },
     {
       "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": 28892.963451409003,
-      "required_pages_per_second_upper_estimate": 0.37060791230755125
+      "required_token_positions_per_second": 28739.32771596405,
+      "required_pages_per_second_upper_estimate": 0.36863723805446375
     }
   ],
   "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": 0.3338729133046172,
-    "producer_p95_seconds": 0.23951535900050658,
-    "gpu_update_seconds": 0.9569111886531566,
-    "producer_batches_per_second": 2.9951516285108917,
-    "gpu_batches_per_second": 1.0450290600191334,
-    "producer_headroom": 2.866094105034795,
+    "producer_batch_seconds": 0.06065465613274057,
+    "producer_p95_seconds": 0.20881386700057192,
+    "gpu_update_seconds": 0.9620266790249988,
+    "producer_batches_per_second": 16.48678046762866,
+    "gpu_batches_per_second": 1.0394722119489312,
+    "producer_headroom": 15.860722661087015,
     "sustainable": true,
-    "buffer_seconds": 60.285404885148864,
+    "buffer_seconds": 60.607680778574924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 543983.5232273676,
-    "upload_bytes_per_second": 2179296.8451801217,
+    "generation_bytes_per_second": 517212.2460824716,
+    "upload_bytes_per_second": 2200623.9726313176,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.72620079627771,
+    "estimated_full_shards_per_hour": 110.98170792441951,
     "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": 1.1295762649997414,
-      "bytes": 7499226,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2592496409997693,
       "bytes": 5842307,
@@ -562,6 +556,12 @@
       "bytes": 896140,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.8745819040013885,
+      "bytes": 13367716,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 659700, "loss": 0.5896466970443726, "lr": 0.0001, "grad_norm": 0.2005715652820951, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96562606454, "tokens_per_second": 153949.98439583473, "data_wait_s": 1.6985999536700547e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 659800, "loss": 0.49065741896629333, "lr": 0.0001, "grad_norm": 0.18866629979146263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96577356469, "tokens_per_second": 153739.58807662022, "data_wait_s": 7.966998964548111e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 659900, "loss": 0.48707088828086853, "lr": 0.0001, "grad_norm": 0.18031609841592164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96591984570, "tokens_per_second": 153111.41243138822, "data_wait_s": 1.3088998457533307e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 660000, "loss": 0.5981903076171875, "lr": 0.0001, "grad_norm": 0.2018008907836487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96606570132, "tokens_per_second": 152092.62098289657, "data_wait_s": 9.361001502838917e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fd27f1fcc0e7f5d2650b1259c03d72359ddd21f444a7859ce67b880f87bb98e
-size 317313
+oid sha256:352d91662c3ee4993d5a3350c3018a57eba06629e245e7ce505d73ac84f11021
+size 319431