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": 76241.70943048918,
-      "required_pages_per_second_upper_estimate": 0.005274578363539892
+      "required_token_positions_per_second": 76344.04720759991,
+      "required_pages_per_second_upper_estimate": 0.00528165832841677
     },
     {
       "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": 9530.213678811147,
-      "required_pages_per_second_upper_estimate": 0.003990101453826035
+      "required_token_positions_per_second": 9543.00590094999,
+      "required_pages_per_second_upper_estimate": 0.003995457290103593
     },
     {
       "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": 19060.427357622295,
-      "required_pages_per_second_upper_estimate": 0.0039697843613210665
+      "required_token_positions_per_second": 19086.01180189998,
+      "required_pages_per_second_upper_estimate": 0.00397511292635694
     },
     {
       "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": 28590.64103643344,
-      "required_pages_per_second_upper_estimate": 0.0017139652747530506
+      "required_token_positions_per_second": 28629.017702849967,
+      "required_pages_per_second_upper_estimate": 0.0017162658973069447
     },
     {
       "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": 28590.64103643344,
-      "required_pages_per_second_upper_estimate": 0.0022227965836431837
+      "required_token_positions_per_second": 28629.017702849967,
+      "required_pages_per_second_upper_estimate": 0.0022257802006559525
     },
     {
       "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": 28590.64103643344,
-      "required_pages_per_second_upper_estimate": 0.36673004497676326
+      "required_token_positions_per_second": 28629.017702849967,
+      "required_pages_per_second_upper_estimate": 0.36722229964790043
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07517905389840962,
-    "producer_p95_seconds": 0.18770330599909357,
-    "gpu_update_seconds": 0.9670297341276042,
-    "producer_batches_per_second": 13.301577342956621,
-    "gpu_batches_per_second": 1.0340943661904456,
-    "producer_headroom": 12.863020801437104,
+    "producer_batch_seconds": 0.05473308060935267,
+    "producer_p95_seconds": 0.18756653099990217,
+    "gpu_update_seconds": 0.9657334487326015,
+    "producer_batches_per_second": 18.27048631041466,
+    "gpu_batches_per_second": 1.035482411127386,
+    "producer_headroom": 17.644419754578532,
     "sustainable": true,
-    "buffer_seconds": 60.92287325003906,
+    "buffer_seconds": 60.841207270153895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317580.31198802224,
-    "upload_bytes_per_second": 4013183.6357162143,
+    "generation_bytes_per_second": 317824.8591670001,
+    "upload_bytes_per_second": 4017866.679639008,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.14534206133366,
+    "estimated_full_shards_per_hour": 68.19781619317534,
     "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": [
@@ -382,6 +382,12 @@
       "bytes": 6845301,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.52379110199945,
+      "bytes": 6381300,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 808100, "loss": 0.5016139149665833, "lr": 3e-05, "grad_norm": 0.27525077283135146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118282451053, "tokens_per_second": 149272.0794929306, "data_wait_s": 1.569300002302043e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808200, "loss": 0.5658459067344666, "lr": 3e-05, "grad_norm": 0.3018091461766138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118296997231, "tokens_per_second": 150670.96458273006, "data_wait_s": 1.4690000170958228e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808300, "loss": 0.5780943036079407, "lr": 3e-05, "grad_norm": 0.3202913211437767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118311574940, "tokens_per_second": 151289.2426151579, "data_wait_s": 9.577999662724324e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 808400, "loss": 0.5267443656921387, "lr": 3e-05, "grad_norm": 0.2612070844248897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118326353095, "tokens_per_second": 132708.07914494834, "data_wait_s": 1.7387999832862988e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b335b33a15ec434dfe32eafa23403573b4a12b94e83afcc2637701c2b0c4f03
-size 176742
+oid sha256:e174488b0a8c5cc496dacbeb7def183efc14a2d1823e87d7f54c2ddc19a1b290
+size 178860