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": 76960.4055129103,
-      "required_pages_per_second_upper_estimate": 0.005324299426126448
+      "required_token_positions_per_second": 76282.90961889723,
+      "required_pages_per_second_upper_estimate": 0.005277428688171569
     },
     {
       "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": 9620.050689113788,
-      "required_pages_per_second_upper_estimate": 0.004027714334029648
+      "required_token_positions_per_second": 9535.363702362154,
+      "required_pages_per_second_upper_estimate": 0.003992257661142121
     },
     {
       "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": 19240.101378227577,
-      "required_pages_per_second_upper_estimate": 0.004007205721490585
+      "required_token_positions_per_second": 19070.727404724308,
+      "required_pages_per_second_upper_estimate": 0.003971929589501907
     },
     {
       "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": 28860.15206734137,
-      "required_pages_per_second_upper_estimate": 0.0017301220495364645
+      "required_token_positions_per_second": 28606.09110708646,
+      "required_pages_per_second_upper_estimate": 0.0017148914828978073
     },
     {
       "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": 28860.15206734137,
-      "required_pages_per_second_upper_estimate": 0.0022437498808425328
+      "required_token_positions_per_second": 28606.09110708646,
+      "required_pages_per_second_upper_estimate": 0.0022239977586787775
     },
     {
       "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": 28860.15206734137,
-      "required_pages_per_second_upper_estimate": 0.370187043102851
+      "required_token_positions_per_second": 28606.09110708646,
+      "required_pages_per_second_upper_estimate": 0.3669282218941068
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0764635192501828,
-    "producer_p95_seconds": 0.27139650100070867,
-    "gpu_update_seconds": 0.9579991101740223,
-    "producer_batches_per_second": 13.078132026961462,
-    "gpu_batches_per_second": 1.0438423056764095,
-    "producer_headroom": 12.528838844567463,
+    "producer_batch_seconds": 0.18765744754676916,
+    "producer_p95_seconds": 0.2599593269987963,
+    "gpu_update_seconds": 0.9665074440440024,
+    "producer_batches_per_second": 5.328858582874915,
+    "gpu_batches_per_second": 1.0346531795097822,
+    "producer_headroom": 5.150381488606379,
     "sustainable": true,
-    "buffer_seconds": 60.353943940963404,
+    "buffer_seconds": 60.889968974772145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294918.05752293556,
-    "upload_bytes_per_second": 4843586.39495015,
+    "generation_bytes_per_second": 330620.2983582574,
+    "upload_bytes_per_second": 4806363.931037319,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.282549803410056,
+    "estimated_full_shards_per_hour": 70.94341958103935,
     "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": 1.2551854249995813,
-      "bytes": 7559192,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.825367109000581,
       "bytes": 1141341,
@@ -562,6 +556,12 @@
       "bytes": 13310041,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.608177831996727,
+      "bytes": 5937731,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 782100, "loss": 0.5047265291213989, "lr": 3e-05, "grad_norm": 0.2702417485551091, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114477135476, "tokens_per_second": 153693.59352186232, "data_wait_s": 1.4005003322381526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782200, "loss": 0.5140717029571533, "lr": 3e-05, "grad_norm": 0.2705502672653411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114491714452, "tokens_per_second": 135976.85138680084, "data_wait_s": 1.3278997357701883e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782300, "loss": 0.5291146636009216, "lr": 3e-05, "grad_norm": 0.24404227322216446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114506464487, "tokens_per_second": 152407.84764650703, "data_wait_s": 2.044500070041977e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 782400, "loss": 0.575973629951477, "lr": 3e-05, "grad_norm": 0.2714590206960609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114521237115, "tokens_per_second": 153492.39907699975, "data_wait_s": 1.1018000805051997e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb9c3aeba33a19338a86c0aaa00f13a79dad41c787235e0b8de4705f99a3ca20
-size 475611
+oid sha256:816e6cb2d39763bea463f35925079a149ca23903fe9ce13a7b7746241c375dc4
+size 477729