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": 75605.45798846474,
-      "required_pages_per_second_upper_estimate": 0.00655328947815873
+      "required_token_positions_per_second": 75845.15831215118,
+      "required_pages_per_second_upper_estimate": 0.006574066094700954
     },
     {
       "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": 9450.682248558092,
-      "required_pages_per_second_upper_estimate": 0.003770018449241301
+      "required_token_positions_per_second": 9480.644789018897,
+      "required_pages_per_second_upper_estimate": 0.003781970954611017
     },
     {
       "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": 18901.364497116185,
-      "required_pages_per_second_upper_estimate": 0.004372498023754179
+      "required_token_positions_per_second": 18961.289578037795,
+      "required_pages_per_second_upper_estimate": 0.00438636063657999
     },
     {
       "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": 28352.046745674277,
-      "required_pages_per_second_upper_estimate": 0.0017264908951184256
+      "required_token_positions_per_second": 28441.934367056692,
+      "required_pages_per_second_upper_estimate": 0.0017319645796567127
     },
     {
       "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": 28352.046745674277,
-      "required_pages_per_second_upper_estimate": 0.002182619041970565
+      "required_token_positions_per_second": 28441.934367056692,
+      "required_pages_per_second_upper_estimate": 0.0021895388398894444
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28352.046745674277,
-      "required_pages_per_second_upper_estimate": 0.3520375323847955
+      "required_token_positions_per_second": 28441.934367056692,
+      "required_pages_per_second_upper_estimate": 0.3531536358078485
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.23956459093764693,
-    "producer_p95_seconds": 0.3229054380026355,
-    "gpu_update_seconds": 0.9751676924071911,
-    "producer_batches_per_second": 4.174239590609101,
-    "gpu_batches_per_second": 1.0254646537063903,
-    "producer_headroom": 4.070583589129015,
+    "producer_batch_seconds": 0.07452981782805068,
+    "producer_p95_seconds": 0.2798562950010819,
+    "gpu_update_seconds": 0.9720857816205258,
+    "producer_batches_per_second": 13.417448601674046,
+    "gpu_batches_per_second": 1.0287157974195853,
+    "producer_headroom": 13.042911011311546,
     "sustainable": true,
-    "buffer_seconds": 60.46039692924585,
+    "buffer_seconds": 60.269318460472604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459201.07362835,
+    "generation_bytes_per_second": 460061.076100161,
     "upload_bytes_per_second": 5535795.565450709,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.53386074674486,
+    "estimated_full_shards_per_hour": 98.71839725736258,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 258500, "loss": 0.6101230606436729, "lr": 0.0003, "grad_norm": 0.14810045976350694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37837180088, "tokens_per_second": 149825.3454071128, "data_wait_s": 1.1959997209487483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258600, "loss": 0.5844476539641619, "lr": 0.0003, "grad_norm": 0.15140719465755104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37851864597, "tokens_per_second": 148721.8763280229, "data_wait_s": 9.61499972618185e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 258700, "loss": 0.5411769356578588, "lr": 0.0003, "grad_norm": 0.1307084377369494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37866358095, "tokens_per_second": 146500.80926690865, "data_wait_s": 7.2500006353948265e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 258800, "loss": 0.5922668986022472, "lr": 0.0003, "grad_norm": 0.1422421201777596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37880753239, "tokens_per_second": 147131.9444011959, "data_wait_s": 1.2044001778122038e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02c9b5cc6e13918a61e421cb50b0096723c4b48b98d93a185b8a2c035c05b94b
-size 460785
+oid sha256:d87ecd5407f37251013c918b00641837f39765ddde0aaa892fa5b223c2db051e
+size 462903