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": 76929.33004858409,
-      "required_pages_per_second_upper_estimate": 0.005322149553399349
+      "required_token_positions_per_second": 77054.38402590342,
+      "required_pages_per_second_upper_estimate": 0.005330801077715505
     },
     {
       "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": 9616.166256073011,
-      "required_pages_per_second_upper_estimate": 0.00402608800303166
+      "required_token_positions_per_second": 9631.798003237927,
+      "required_pages_per_second_upper_estimate": 0.0040326326891416105
     },
     {
       "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": 19232.332512146022,
-      "required_pages_per_second_upper_estimate": 0.004005587671564573
+      "required_token_positions_per_second": 19263.596006475855,
+      "required_pages_per_second_upper_estimate": 0.004012099032959688
     },
     {
       "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": 28848.498768219033,
-      "required_pages_per_second_upper_estimate": 0.001729423452047638
+      "required_token_positions_per_second": 28895.394009713782,
+      "required_pages_per_second_upper_estimate": 0.0017322347501703605
     },
     {
       "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": 28848.498768219033,
-      "required_pages_per_second_upper_estimate": 0.002242843888093217
+      "required_token_positions_per_second": 28895.394009713782,
+      "required_pages_per_second_upper_estimate": 0.002246489786849066
     },
     {
       "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": 28848.498768219033,
-      "required_pages_per_second_upper_estimate": 0.37003756709404745
+      "required_token_positions_per_second": 28895.394009713782,
+      "required_pages_per_second_upper_estimate": 0.3706390888997548
     }
   ],
   "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": 0.16491479917181096,
-    "producer_p95_seconds": 0.34410336799919605,
-    "gpu_update_seconds": 0.9583860921892559,
-    "producer_batches_per_second": 6.063737184424446,
-    "gpu_batches_per_second": 1.0434208177162556,
-    "producer_headroom": 5.811401384243227,
+    "producer_batch_seconds": 0.09654186317163749,
+    "producer_p95_seconds": 0.3272545789950527,
+    "gpu_update_seconds": 0.9568306973320923,
+    "producer_batches_per_second": 10.358200755066685,
+    "gpu_batches_per_second": 1.0451169708374488,
+    "producer_headroom": 9.911044451576261,
     "sustainable": true,
-    "buffer_seconds": 60.37832380792312,
+    "buffer_seconds": 60.28033393192181,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244531.29408867637,
-    "upload_bytes_per_second": 3737474.822197447,
+    "generation_bytes_per_second": 244352.84151134157,
+    "upload_bytes_per_second": 3734871.8411563905,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.47072331423968,
+    "estimated_full_shards_per_hour": 52.43243154530702,
     "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": 1.6460893040020892,
-      "bytes": 7447219,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 0.9642833869984315,
       "bytes": 493177,
@@ -562,6 +556,12 @@
       "bytes": 318500,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 1.7101979840008426,
+      "bytes": 7394765,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 880000, "loss": 0.5373256802558899, "lr": 3e-05, "grad_norm": 0.3214243903464673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128807500728, "tokens_per_second": 153504.90885856643, "data_wait_s": 1.5455996617674828e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880100, "loss": 0.5131340622901917, "lr": 3e-05, "grad_norm": 0.31855802461413985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128822017948, "tokens_per_second": 136166.5131094943, "data_wait_s": 1.3425000361166894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880200, "loss": 0.5502160787582397, "lr": 3e-05, "grad_norm": 0.33095797507033536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128836474938, "tokens_per_second": 150345.95068226027, "data_wait_s": 1.3336997653823346e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 880300, "loss": 0.6008081436157227, "lr": 3e-05, "grad_norm": 0.31065557838755636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128850995000, "tokens_per_second": 151393.7339445974, "data_wait_s": 9.249000868294388e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d027d0c22608db3afc767421ce5e821a57d8167941f74ce5093484fd49cdcc06
-size 804915
+oid sha256:a214198ede06130f90cd6514a3e2e9576b4d38ec51fd82dab04778d9234f38f6
+size 807033