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": 76142.93958208182,
-      "required_pages_per_second_upper_estimate": 0.005267745236249448
+      "required_token_positions_per_second": 76044.2173892861,
+      "required_pages_per_second_upper_estimate": 0.005260915405884793
     },
     {
       "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": 9517.867447760227,
-      "required_pages_per_second_upper_estimate": 0.003984932344703637
+      "required_token_positions_per_second": 9505.527173660763,
+      "required_pages_per_second_upper_estimate": 0.0039797657296324175
     },
     {
       "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": 19035.734895520454,
-      "required_pages_per_second_upper_estimate": 0.003964641572649269
+      "required_token_positions_per_second": 19011.054347321526,
+      "required_pages_per_second_upper_estimate": 0.003959501265329257
     },
     {
       "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": 28553.60234328068,
-      "required_pages_per_second_upper_estimate": 0.0017117448616533525
+      "required_token_positions_per_second": 28516.581520982287,
+      "required_pages_per_second_upper_estimate": 0.0017095255198840865
     },
     {
       "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": 28553.60234328068,
-      "required_pages_per_second_upper_estimate": 0.0022199169881665505
+      "required_token_positions_per_second": 28516.581520982287,
+      "required_pages_per_second_upper_estimate": 0.0022170387820702386
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28553.60234328068,
-      "required_pages_per_second_upper_estimate": 0.3662549523900499
+      "required_token_positions_per_second": 28516.581520982287,
+      "required_pages_per_second_upper_estimate": 0.36578008903146814
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4665424931565667,
-    "producer_p95_seconds": 0.34825134000129765,
-    "gpu_update_seconds": 0.968284129883395,
-    "producer_batches_per_second": 2.1434274791008385,
-    "gpu_batches_per_second": 1.032754714383705,
-    "producer_headroom": 2.0754468115693143,
+    "producer_batch_seconds": 0.09913348502357167,
+    "producer_p95_seconds": 0.35871307799970964,
+    "gpu_update_seconds": 0.9695411765837644,
+    "producer_batches_per_second": 10.08740890892944,
+    "gpu_batches_per_second": 1.0314157089475653,
+    "producer_headroom": 9.780158302244995,
     "sustainable": true,
-    "buffer_seconds": 60.03361605277049,
+    "buffer_seconds": 60.11155294819339,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287966.9278012241,
-    "upload_bytes_per_second": 4744654.739089099,
+    "generation_bytes_per_second": 297429.7054266224,
+    "upload_bytes_per_second": 4696556.373573309,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.79099917914908,
+    "estimated_full_shards_per_hour": 63.82149097537045,
     "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": 1.128546699997969,
-      "bytes": 5952550,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.3368553659965983,
       "bytes": 12766871,
@@ -562,6 +556,12 @@
       "bytes": 7243506,
       "pages": 24,
       "shards": 3
+    },
+    {
+      "seconds": 2.0522459450003225,
+      "bytes": 5876405,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -427,3 +427,4 @@
 {"step": 687700, "loss": 0.5363231897354126, "lr": 0.0001, "grad_norm": 0.21060351684668194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100660878824, "tokens_per_second": 153150.24595320888, "data_wait_s": 2.3993998183868825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687800, "loss": 0.5660990476608276, "lr": 0.0001, "grad_norm": 0.2114553653710195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100675435495, "tokens_per_second": 137738.63495659753, "data_wait_s": 1.430800330126658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687900, "loss": 0.5393136739730835, "lr": 0.0001, "grad_norm": 0.23868798620330395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100690033392, "tokens_per_second": 150711.7483638031, "data_wait_s": 1.4796998584643006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688000, "loss": 0.4782137870788574, "lr": 0.0001, "grad_norm": 0.20103014709600134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100704686333, "tokens_per_second": 151602.26420412486, "data_wait_s": 1.779800368240103e-05, "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 sha256:7379613fc1df28e4a65c3adf7e9b956afce50a38674627006a4c01016ffa2b83
-size 915051
+oid sha256:948ad680ecccd11a61fbd23f75011530aae3b536fdf90ea117f3b83333fc86fd
+size 917169