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": 77387.11529998193,
-      "required_pages_per_second_upper_estimate": 0.005353820199299178
+      "required_token_positions_per_second": 77148.37888219004,
+      "required_pages_per_second_upper_estimate": 0.005337303859971527
     },
     {
       "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": 9673.389412497741,
-      "required_pages_per_second_upper_estimate": 0.004050046143671306
+      "required_token_positions_per_second": 9643.547360273755,
+      "required_pages_per_second_upper_estimate": 0.0040375518995780365
     },
     {
       "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": 19346.778824995483,
-      "required_pages_per_second_upper_estimate": 0.004029423820378884
+      "required_token_positions_per_second": 19287.09472054751,
+      "required_pages_per_second_upper_estimate": 0.00401699319539805
     },
     {
       "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": 29020.168237493224,
-      "required_pages_per_second_upper_estimate": 0.0017397147746065237
+      "required_token_positions_per_second": 28930.642080821264,
+      "required_pages_per_second_upper_estimate": 0.0017343478182125659
     },
     {
       "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": 29020.168237493224,
-      "required_pages_per_second_upper_estimate": 0.002256190434235091
+      "required_token_positions_per_second": 28930.642080821264,
+      "required_pages_per_second_upper_estimate": 0.002249230169337793
     },
     {
       "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": 29020.168237493224,
-      "required_pages_per_second_upper_estimate": 0.3722395587215816
+      "required_token_positions_per_second": 28930.642080821264,
+      "required_pages_per_second_upper_estimate": 0.37109121330949146
     }
   ],
   "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": 0.11072897771090595,
-    "producer_p95_seconds": 0.24503438600004301,
-    "gpu_update_seconds": 0.9527167373302674,
-    "producer_batches_per_second": 9.031059625700019,
-    "gpu_batches_per_second": 1.0496299275713694,
-    "producer_headroom": 8.604041661232028,
+    "producer_batch_seconds": 0.0775343519296996,
+    "producer_p95_seconds": 0.2292477430000872,
+    "gpu_update_seconds": 0.9556649286511496,
+    "producer_batches_per_second": 12.897508976494704,
+    "gpu_batches_per_second": 1.046391857668593,
+    "producer_headroom": 12.325696995799373,
     "sustainable": true,
-    "buffer_seconds": 60.021154451806844,
+    "buffer_seconds": 60.20689050502242,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348715.41331840673,
-    "upload_bytes_per_second": 5236330.361118066,
+    "generation_bytes_per_second": 353125.30569570774,
+    "upload_bytes_per_second": 5300138.259366258,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.82621001876976,
+    "estimated_full_shards_per_hour": 75.77247026589798,
     "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": 1.2937560119999034,
-      "bytes": 947666,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.5761228700002903,
       "bytes": 18281686,
@@ -562,6 +556,12 @@
       "bytes": 2754086,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.24713432200042,
+      "bytes": 5975341,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -152,3 +152,4 @@
 {"step": 900200, "loss": 0.5399182438850403, "lr": 3e-05, "grad_norm": 0.3283343396842356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131763803760, "tokens_per_second": 153995.44107381676, "data_wait_s": 1.1669999366858974e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 900300, "loss": 0.5195290446281433, "lr": 3e-05, "grad_norm": 0.30953746353027317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131778484935, "tokens_per_second": 153178.59357082617, "data_wait_s": 2.0464000044739805e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 900400, "loss": 0.554978609085083, "lr": 3e-05, "grad_norm": 0.32366369021997676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131793093122, "tokens_per_second": 151810.67255527276, "data_wait_s": 1.5450999853783287e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 900500, "loss": 0.577768862247467, "lr": 3e-05, "grad_norm": 0.3444207151321536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131807727087, "tokens_per_second": 151676.22166932866, "data_wait_s": 1.2712000170722604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dac0d117ca852232e16b170d6bc20eb44e532617fc9250edcd67b0c268316d69
-size 328686
+oid sha256:0b5b4ba1f01e2eba7ec68fc3a9b7f5c4367d34ed59ed998465f225bf3b3ca3d9
+size 330804