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": 77034.63253554079,
-      "required_pages_per_second_upper_estimate": 0.005329434623782446
+      "required_token_positions_per_second": 76855.4172462193,
+      "required_pages_per_second_upper_estimate": 0.005317036094230423
     },
     {
       "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": 9629.329066942599,
-      "required_pages_per_second_upper_estimate": 0.004031598997072009
+      "required_token_positions_per_second": 9606.927155777412,
+      "required_pages_per_second_upper_estimate": 0.004022219784672246
     },
     {
       "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": 19258.658133885197,
-      "required_pages_per_second_upper_estimate": 0.004011070604319511
+      "required_token_positions_per_second": 19213.854311554824,
+      "required_pages_per_second_upper_estimate": 0.0040017391496843455
     },
     {
       "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": 28887.987200827796,
-      "required_pages_per_second_upper_estimate": 0.001731790723287188
+      "required_token_positions_per_second": 28820.781467332235,
+      "required_pages_per_second_upper_estimate": 0.0017277618421813424
     },
     {
       "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": 28887.987200827796,
-      "required_pages_per_second_upper_estimate": 0.0022459139400372898
+      "required_token_positions_per_second": 28820.781467332235,
+      "required_pages_per_second_upper_estimate": 0.0022406889898647906
     },
     {
       "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": 28887.987200827796,
-      "required_pages_per_second_upper_estimate": 0.3705440823081771
+      "required_token_positions_per_second": 28820.781467332235,
+      "required_pages_per_second_upper_estimate": 0.3696820393187906
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2693688793124913,
-    "producer_p95_seconds": 0.1534350709998762,
-    "gpu_update_seconds": 0.9570760263701493,
-    "producer_batches_per_second": 3.712381335781232,
-    "gpu_batches_per_second": 1.0448490741040146,
-    "producer_headroom": 3.5530311772202086,
+    "producer_batch_seconds": 0.06807713191406961,
+    "producer_p95_seconds": 0.15511187999982212,
+    "gpu_update_seconds": 0.9593077839106633,
+    "producer_batches_per_second": 14.689220475125927,
+    "gpu_batches_per_second": 1.0424183111737642,
+    "producer_headroom": 14.091483541368193,
     "sustainable": true,
-    "buffer_seconds": 60.29578966131941,
+    "buffer_seconds": 60.43639038637179,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 74,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526905.4818317355,
-    "upload_bytes_per_second": 4240662.542883857,
+    "generation_bytes_per_second": 545190.2699881681,
+    "upload_bytes_per_second": 4175318.8961378178,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.06165066923187,
+    "estimated_full_shards_per_hour": 116.98514055951865,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 17682755,
       "pages": 37,
       "shards": 5
+    },
+    {
+      "seconds": 0.9498594410006262,
+      "bytes": 1178438,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 443500, "loss": 0.5982214212417603, "lr": 0.0003, "grad_norm": 0.16534686805696577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64915146340, "tokens_per_second": 150582.5599688112, "data_wait_s": 1.2893000075564487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443600, "loss": 0.5966102480888367, "lr": 0.0003, "grad_norm": 0.16662841326636885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64929773473, "tokens_per_second": 152271.8874992761, "data_wait_s": 1.6013999811548274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443700, "loss": 0.5953665971755981, "lr": 0.0003, "grad_norm": 0.15147647180804655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64944491492, "tokens_per_second": 152969.85481276517, "data_wait_s": 1.2269999842828838e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 443800, "loss": 0.5933117270469666, "lr": 0.0003, "grad_norm": 0.17910241468328222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64959227114, "tokens_per_second": 153051.33207088735, "data_wait_s": 1.30259995785309e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f3438ee29af86f8ecd2ef8aff324cfa7d538cf5ce116a3f663832213313e7ee
-size 74295
+oid sha256:3bc47bf4e826bd061fc0aa6e684b503522765fb4e4729d790327253448380daf
+size 76413