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": 75949.53481050153,
-      "required_pages_per_second_upper_estimate": 0.006583113185038146
+      "required_token_positions_per_second": 75001.57716008804,
+      "required_pages_per_second_upper_estimate": 0.0065009466184770045
     },
     {
       "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": 9493.691851312691,
-      "required_pages_per_second_upper_estimate": 0.0037871756228309763
+      "required_token_positions_per_second": 9375.197145011005,
+      "required_pages_per_second_upper_estimate": 0.003739906312833495
     },
     {
       "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": 18987.383702625382,
-      "required_pages_per_second_upper_estimate": 0.004392397053062421
+      "required_token_positions_per_second": 18750.39429002201,
+      "required_pages_per_second_upper_estimate": 0.004337573723327839
     },
     {
       "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": 28481.075553938073,
-      "required_pages_per_second_upper_estimate": 0.0017343480725798522
+      "required_token_positions_per_second": 28125.591435033017,
+      "required_pages_per_second_upper_estimate": 0.001712700954819565
     },
     {
       "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": 28481.075553938073,
-      "required_pages_per_second_upper_estimate": 0.002192552036805313
+      "required_token_positions_per_second": 28125.591435033017,
+      "required_pages_per_second_upper_estimate": 0.0021651858852889793
     },
     {
       "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": 28481.075553938073,
-      "required_pages_per_second_upper_estimate": 0.3536396383517895
+      "required_token_positions_per_second": 28125.591435033017,
+      "required_pages_per_second_upper_estimate": 0.3492257153238017
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2537207677266622,
-    "producer_p95_seconds": 0.2026793990007718,
-    "gpu_update_seconds": 0.9707498562559417,
-    "producer_batches_per_second": 3.9413407462069383,
-    "gpu_batches_per_second": 1.030131494283061,
-    "producer_headroom": 3.826055962836071,
+    "producer_batch_seconds": 0.08051134822679273,
+    "producer_p95_seconds": 0.21015008699760074,
+    "gpu_update_seconds": 0.9830193282819955,
+    "producer_batches_per_second": 12.420609293277465,
+    "gpu_batches_per_second": 1.0172739957694232,
+    "producer_headroom": 12.209699004330725,
     "sustainable": true,
-    "buffer_seconds": 60.18649108786838,
+    "buffer_seconds": 60.94719835348372,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 568802.4177024852,
-    "upload_bytes_per_second": 5910912.288721966,
+    "generation_bytes_per_second": 568540.218189866,
+    "upload_bytes_per_second": 5908525.30705968,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.02587889817199,
+    "estimated_full_shards_per_hour": 60.997747942314085,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 6556698,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0323645409989695,
+      "bytes": 5955575,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 96300, "loss": 0.603190841153264, "lr": 0.0003, "grad_norm": 0.13281830211263865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14092219734, "tokens_per_second": 149040.19338791494, "data_wait_s": 1.1356001778040081e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96400, "loss": 0.6376859731972218, "lr": 0.0003, "grad_norm": 0.13875480605009713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14106662294, "tokens_per_second": 147246.16452948147, "data_wait_s": 1.3895001757191494e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96500, "loss": 0.6454388238489628, "lr": 0.0003, "grad_norm": 0.1403733392987137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14121248610, "tokens_per_second": 149122.40453180103, "data_wait_s": 1.3905999367125332e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 96600, "loss": 0.6495152413845062, "lr": 0.0003, "grad_norm": 0.14796350184121645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14135810183, "tokens_per_second": 148593.40359386528, "data_wait_s": 1.0016003216151148e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9069b29f19f579dfdd95f5ac58a11eafb48f9dc14b9ce098939c919d4710d5c
-size 245301
+oid sha256:f82e6c5d6e8abe106453834f8f5b92add8db559326c35c67e9905ab71dce2517
+size 247419