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": 75150.54189164305,
-      "required_pages_per_second_upper_estimate": 0.006513858503860532
+      "required_token_positions_per_second": 75426.85732264986,
+      "required_pages_per_second_upper_estimate": 0.006537808825105145
     },
     {
       "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": 9393.817736455381,
-      "required_pages_per_second_upper_estimate": 0.0037473343451633083
+      "required_token_positions_per_second": 9428.357165331232,
+      "required_pages_per_second_upper_estimate": 0.003761112639752366
     },
     {
       "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": 18787.635472910762,
-      "required_pages_per_second_upper_estimate": 0.004346188815566718
+      "required_token_positions_per_second": 18856.714330662464,
+      "required_pages_per_second_upper_estimate": 0.004362168993561202
     },
     {
       "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": 28181.453209366144,
-      "required_pages_per_second_upper_estimate": 0.0017161026437923733
+      "required_token_positions_per_second": 28285.071495993696,
+      "required_pages_per_second_upper_estimate": 0.001722412453804856
     },
     {
       "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": 28181.453209366144,
-      "required_pages_per_second_upper_estimate": 0.0021694862793124317
+      "required_token_positions_per_second": 28285.071495993696,
+      "required_pages_per_second_upper_estimate": 0.002177463101850796
     },
     {
       "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": 28181.453209366144,
-      "required_pages_per_second_upper_estimate": 0.34991933160368704
+      "required_token_positions_per_second": 28285.071495993696,
+      "required_pages_per_second_upper_estimate": 0.35120592393550415
     }
   ],
   "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": 0.0709647496796535,
-    "producer_p95_seconds": 0.18526915700022073,
-    "gpu_update_seconds": 0.981070770005968,
-    "producer_batches_per_second": 14.091503239483883,
-    "gpu_batches_per_second": 1.0192944592508009,
-    "producer_headroom": 13.824761933702044,
+    "producer_batch_seconds": 0.05208513217183253,
+    "producer_p95_seconds": 0.18587464800020825,
+    "gpu_update_seconds": 0.9774767585054919,
+    "producer_batches_per_second": 19.199336899077636,
+    "gpu_batches_per_second": 1.0230422271409756,
+    "producer_headroom": 18.76690559756529,
     "sustainable": true,
-    "buffer_seconds": 60.826387740370016,
+    "buffer_seconds": 60.603559027340495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394478.5167755432,
-    "upload_bytes_per_second": 3078760.5405726614,
+    "generation_bytes_per_second": 393534.9287613518,
+    "upload_bytes_per_second": 3066870.8165609036,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.64590671014521,
+    "estimated_full_shards_per_hour": 84.4434346879045,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 6211390,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.7338976749997528,
+      "bytes": 1218142,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -49,3 +49,4 @@
 {"step": 242000, "loss": 0.6067950278520584, "lr": 0.0003, "grad_norm": 0.14927075253257158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35421246671, "tokens_per_second": 149102.67931153878, "data_wait_s": 1.817799966374878e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242100, "loss": 0.6696290075778961, "lr": 0.0003, "grad_norm": 0.17687013251592487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35435978706, "tokens_per_second": 149735.172557017, "data_wait_s": 1.2118000086047687e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242200, "loss": 0.6197372898459435, "lr": 0.0003, "grad_norm": 0.1444105900824176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35450713692, "tokens_per_second": 150344.98341179563, "data_wait_s": 1.4904000636306591e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 242300, "loss": 0.6313554421067238, "lr": 0.0003, "grad_norm": 0.158225694374267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35465399501, "tokens_per_second": 149949.53320490653, "data_wait_s": 1.2076000530214515e-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:bd603af2fb6cf5f44d0bfaa62f4b78403cd0900af79f192c05c9ff0ce6256a65
-size 108966
+oid sha256:6074d0a465eea51aeeccd9bec4080aa0633efe5ca2935e5daa8b267c722b48d1
+size 111084