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": 74720.6691967026,
-      "required_pages_per_second_upper_estimate": 0.006476598228165478
+      "required_token_positions_per_second": 73943.86823084886,
+      "required_pages_per_second_upper_estimate": 0.006409267089229325
     },
     {
       "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": 9340.083649587825,
-      "required_pages_per_second_upper_estimate": 0.0037258990145156473
+      "required_token_positions_per_second": 9242.983528856108,
+      "required_pages_per_second_upper_estimate": 0.003687164324579587
     },
     {
       "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": 18680.16729917565,
-      "required_pages_per_second_upper_estimate": 0.00432132794494836
+      "required_token_positions_per_second": 18485.967057712216,
+      "required_pages_per_second_upper_estimate": 0.004276403136893312
     },
     {
       "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": 28020.250948763478,
-      "required_pages_per_second_upper_estimate": 0.0017062862718845682
+      "required_token_positions_per_second": 27728.950586568324,
+      "required_pages_per_second_upper_estimate": 0.0016885476081617678
     },
     {
       "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": 28020.250948763478,
-      "required_pages_per_second_upper_estimate": 0.002157076483054821
+      "required_token_positions_per_second": 27728.950586568324,
+      "required_pages_per_second_upper_estimate": 0.0021346513748020253
     },
     {
       "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": 28020.250948763478,
-      "required_pages_per_second_upper_estimate": 0.34791773903626255
+      "required_token_positions_per_second": 27728.950586568324,
+      "required_pages_per_second_upper_estimate": 0.3443007634573963
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11888074760156542,
-    "producer_p95_seconds": 0.47066094200272346,
-    "gpu_update_seconds": 0.9867149316598143,
-    "producer_batches_per_second": 8.411790976883392,
-    "gpu_batches_per_second": 1.01346393767229,
-    "producer_headroom": 8.300039758892138,
+    "producer_batch_seconds": 0.198605080398238,
+    "producer_p95_seconds": 0.46470169399981387,
+    "gpu_update_seconds": 0.997080647307023,
+    "producer_batches_per_second": 5.035117923442969,
+    "gpu_batches_per_second": 1.0029279002665048,
+    "producer_headroom": 5.020418638373709,
     "sustainable": true,
-    "buffer_seconds": 60.18961083124867,
+    "buffer_seconds": 60.8219194857284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368810.7279025855,
-    "upload_bytes_per_second": 4378923.702938631,
+    "generation_bytes_per_second": 364879.59186824976,
+    "upload_bytes_per_second": 4436585.254587113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.13819673355269,
+    "estimated_full_shards_per_hour": 78.29466645274753,
     "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": 1.1135386019959697,
-      "bytes": 1273006,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2612176139955409,
       "bytes": 7427910,
@@ -562,6 +556,12 @@
       "bytes": 12112389,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.0702350749998004,
+      "bytes": 6590156,
+      "pages": 42,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 278100, "loss": 0.5916734673082829, "lr": 0.0003, "grad_norm": 0.14871018837106634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40706113567, "tokens_per_second": 146715.18309275972, "data_wait_s": 1.917599729495123e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278200, "loss": 0.6078765690326691, "lr": 0.0003, "grad_norm": 0.1374363847299842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40720756769, "tokens_per_second": 146157.92697024057, "data_wait_s": 1.7257996660191566e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278300, "loss": 0.6173438597470522, "lr": 0.0003, "grad_norm": 0.17170804097674056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40735477528, "tokens_per_second": 146738.47210259954, "data_wait_s": 2.0571998902596533e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278400, "loss": 0.6388246286660433, "lr": 0.0003, "grad_norm": 0.16316806816724105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40750257836, "tokens_per_second": 147282.42974894305, "data_wait_s": 1.0408999514766037e-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 sha256:67bf5a65a99612bfbdb3dc7b35f9f25be820077aa26bc703aa3a380a3609b877
-size 879045
+oid sha256:eed29853c1d84d41faa6cfcdfb22b2ab21b2579dcc1436af2bcace918810b48f
+size 881163