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": 76850.32401609197,
-      "required_pages_per_second_upper_estimate": 0.005316683732752292
+      "required_token_positions_per_second": 76899.38131576907,
+      "required_pages_per_second_upper_estimate": 0.005320077630572572
     },
     {
       "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": 9606.290502011496,
-      "required_pages_per_second_upper_estimate": 0.004021953231035299
+      "required_token_positions_per_second": 9612.422664471134,
+      "required_pages_per_second_upper_estimate": 0.004024520639403037
     },
     {
       "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": 19212.58100402299,
-      "required_pages_per_second_upper_estimate": 0.004001473953304842
+      "required_token_positions_per_second": 19224.84532894227,
+      "required_pages_per_second_upper_estimate": 0.00400402828875353
     },
     {
       "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": 28818.871506034488,
-      "required_pages_per_second_upper_estimate": 0.0017276473429178845
+      "required_token_positions_per_second": 28837.267993413403,
+      "required_pages_per_second_upper_estimate": 0.0017287501842464406
     },
     {
       "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": 28818.871506034488,
-      "required_pages_per_second_upper_estimate": 0.002240540498774923
+      "required_token_positions_per_second": 28837.267993413403,
+      "required_pages_per_second_upper_estimate": 0.002241970744751037
     },
     {
       "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": 28818.871506034488,
-      "required_pages_per_second_upper_estimate": 0.3696575403860198
+      "required_token_positions_per_second": 28837.267993413403,
+      "required_pages_per_second_upper_estimate": 0.3698935107735073
     }
   ],
   "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": 0.4611673378906431,
-    "producer_p95_seconds": 0.20829420399968512,
-    "gpu_update_seconds": 0.9593713617207629,
-    "producer_batches_per_second": 2.16841028806149,
-    "gpu_batches_per_second": 1.0423492298189558,
-    "producer_headroom": 2.0803107308268634,
+    "producer_batch_seconds": 0.09294128937497703,
+    "producer_p95_seconds": 0.2145475670004089,
+    "gpu_update_seconds": 0.9587593390023965,
+    "producer_batches_per_second": 10.75948060033299,
+    "gpu_batches_per_second": 1.0430146120302881,
+    "producer_headroom": 10.315752508384366,
     "sustainable": true,
-    "buffer_seconds": 60.44039578840806,
+    "buffer_seconds": 60.40183835715098,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258442.85343619043,
-    "upload_bytes_per_second": 3369852.9986776123,
+    "generation_bytes_per_second": 268617.8734574305,
+    "upload_bytes_per_second": 3408176.419721202,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.4558201056889,
+    "estimated_full_shards_per_hour": 57.63914253990351,
     "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": 1.107387490000292,
-      "bytes": 5957600,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 25.62068473599993,
       "bytes": 7109407,
@@ -562,6 +556,12 @@
       "bytes": 999510,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.372887202000129,
+      "bytes": 11719387,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 698200, "loss": 0.5292916893959045, "lr": 8e-05, "grad_norm": 0.20238014899588047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102197454638, "tokens_per_second": 152919.03727053688, "data_wait_s": 9.194998710881919e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 698300, "loss": 0.5354017019271851, "lr": 8e-05, "grad_norm": 0.23375011422462638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102212240388, "tokens_per_second": 145206.8832564163, "data_wait_s": 1.553699985379353e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698400, "loss": 0.5343481302261353, "lr": 8e-05, "grad_norm": 0.2049336983288719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102226991901, "tokens_per_second": 153268.80534206703, "data_wait_s": 1.8017000911640935e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 698500, "loss": 0.5305786728858948, "lr": 8e-05, "grad_norm": 0.21394717140636219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102241679933, "tokens_per_second": 153088.39200416883, "data_wait_s": 1.3371998647926375e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a4fef3f744db78554e773da665da5d3a36f199ddc6e36f12d557e16362fae22
-size 285543
+oid sha256:6e085b2af1fa38819eadb12e93e4b8721ca43884323c3f322177cfe11c04cb05
+size 287661