Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74863.04952718689,
-      "required_pages_per_second_upper_estimate": 0.006488939394352208
+      "required_token_positions_per_second": 74182.17834978638,
+      "required_pages_per_second_upper_estimate": 0.006429923206347355
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9357.88119089836,
-      "required_pages_per_second_upper_estimate": 0.0037329987198413757
+      "required_token_positions_per_second": 9272.772293723297,
+      "required_pages_per_second_upper_estimate": 0.003699047508266833
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18715.76238179672,
-      "required_pages_per_second_upper_estimate": 0.004329562240860687
+      "required_token_positions_per_second": 18545.544587446595,
+      "required_pages_per_second_upper_estimate": 0.004290185349868648
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28073.64357269508,
-      "required_pages_per_second_upper_estimate": 0.0017095376025525546
+      "required_token_positions_per_second": 27818.31688116989,
+      "required_pages_per_second_upper_estimate": 0.0016939895466342916
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28073.64357269508,
-      "required_pages_per_second_upper_estimate": 0.002161186795045318
+      "required_token_positions_per_second": 27818.31688116989,
+      "required_pages_per_second_upper_estimate": 0.00214153103954219
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28073.64357269508,
-      "required_pages_per_second_upper_estimate": 0.3485806967318758
+      "required_token_positions_per_second": 27818.31688116989,
+      "required_pages_per_second_upper_estimate": 0.34541039374659954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08557245334364438,
-    "producer_p95_seconds": 0.21060984399991867,
-    "gpu_update_seconds": 0.9848383209827074,
-    "producer_batches_per_second": 11.686003625304167,
-    "gpu_batches_per_second": 1.01539509449852,
-    "producer_headroom": 11.508824189342388,
+    "producer_batch_seconds": 0.374912965593694,
+    "producer_p95_seconds": 0.2547503229998256,
+    "gpu_update_seconds": 0.993877527461585,
+    "producer_batches_per_second": 2.667285721677959,
+    "gpu_batches_per_second": 1.0061601881210174,
+    "producer_headroom": 2.6509553380948794,
     "sustainable": true,
-    "buffer_seconds": 60.07513757994515,
+    "buffer_seconds": 60.62652917515668,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 470599.97972507915,
+    "generation_bytes_per_second": 427359.5393192017,
     "upload_bytes_per_second": 5893495.417836866,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.97980064214974,
+    "estimated_full_shards_per_hour": 91.70140871698416,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 125900, "loss": 0.6171390563249588, "lr": 0.0003, "grad_norm": 0.14805788000001513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18424806416, "tokens_per_second": 150105.81968424015, "data_wait_s": 1.267200059373863e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126000, "loss": 0.6136645209044218, "lr": 0.0003, "grad_norm": 0.13043100638691812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18439450563, "tokens_per_second": 149323.9691268889, "data_wait_s": 1.714799873298034e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126100, "loss": 0.5676320344209671, "lr": 0.0003, "grad_norm": 0.12993174176535047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18454002821, "tokens_per_second": 147875.12192090612, "data_wait_s": 1.135099955718033e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 126200, "loss": 0.5653232540935278, "lr": 0.0003, "grad_norm": 0.1336714102744638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18468457288, "tokens_per_second": 147792.7347952092, "data_wait_s": 1.3756998669123277e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcf4b7badc4a7951b8a5f3d15289457ec4a0922a2bea55b6f87f9bccc0268a07
-size 236829
+oid sha256:748791c8094fec11769dfc74c89839c45642a7700025856375f52a4b01497dd8
+size 238947