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": 77670.81735698314,
-      "required_pages_per_second_upper_estimate": 0.005373447365881988
+      "required_token_positions_per_second": 77548.71689394301,
+      "required_pages_per_second_upper_estimate": 0.005365000172536764
     },
     {
       "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": 9708.852169622893,
-      "required_pages_per_second_upper_estimate": 0.004064893659533027
+      "required_token_positions_per_second": 9693.589611742877,
+      "required_pages_per_second_upper_estimate": 0.004058503545267116
     },
     {
       "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": 19417.704339245785,
-      "required_pages_per_second_upper_estimate": 0.00404419573456563
+      "required_token_positions_per_second": 19387.179223485753,
+      "required_pages_per_second_upper_estimate": 0.004037838157954251
     },
     {
       "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": 29126.556508868674,
-      "required_pages_per_second_upper_estimate": 0.0017460925890299974
+      "required_token_positions_per_second": 29080.76883522863,
+      "required_pages_per_second_upper_estimate": 0.0017433476879090064
     },
     {
       "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": 29126.556508868674,
-      "required_pages_per_second_upper_estimate": 0.002264461654381981
+      "required_token_positions_per_second": 29080.76883522863,
+      "required_pages_per_second_upper_estimate": 0.00226090186415516
     },
     {
       "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": 29126.556508868674,
-      "required_pages_per_second_upper_estimate": 0.37360419323595995
+      "required_token_positions_per_second": 29080.76883522863,
+      "required_pages_per_second_upper_estimate": 0.3730168781214791
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09424460097659448,
-    "producer_p95_seconds": 0.2267190339989611,
-    "gpu_update_seconds": 0.9492368241876284,
-    "producer_batches_per_second": 10.610687398934912,
-    "gpu_batches_per_second": 1.0534778829885951,
-    "producer_headroom": 10.072055209012664,
+    "producer_batch_seconds": 0.4502592875782341,
+    "producer_p95_seconds": 0.22850390899839113,
+    "gpu_update_seconds": 0.9507313976687932,
+    "producer_batches_per_second": 2.2209425270905636,
+    "gpu_batches_per_second": 1.0518217894686281,
+    "producer_headroom": 2.111519792922873,
     "sustainable": true,
-    "buffer_seconds": 60.75115674800822,
+    "buffer_seconds": 60.846809450802766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348776.45460975275,
-    "upload_bytes_per_second": 5146947.74127454,
+    "generation_bytes_per_second": 298338.4820591867,
+    "upload_bytes_per_second": 5092014.01384785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.83930805892408,
+    "estimated_full_shards_per_hour": 64.01649328548146,
     "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": 1.2208401020016026,
-      "bytes": 11777262,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9635194839993346,
       "bytes": 1029009,
@@ -562,6 +556,12 @@
       "bytes": 2586599,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9690040009991208,
+      "bytes": 5902816,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -183,3 +183,4 @@
 {"step": 575400, "loss": 0.5910930037498474, "lr": 0.0003, "grad_norm": 0.17306087163264713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84221195839, "tokens_per_second": 151026.81385286932, "data_wait_s": 1.5139998140512034e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575500, "loss": 0.561769962310791, "lr": 0.0003, "grad_norm": 0.15479443685832425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84235675989, "tokens_per_second": 151186.59097593135, "data_wait_s": 1.3323999155545607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575600, "loss": 0.5846127271652222, "lr": 0.0003, "grad_norm": 0.1487987002422767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84250301136, "tokens_per_second": 131601.4857546011, "data_wait_s": 1.150900061475113e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 575700, "loss": 0.5894057154655457, "lr": 0.0003, "grad_norm": 0.15861235980839236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84265001282, "tokens_per_second": 67196.43554143806, "data_wait_s": 7.2690017987042665e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:589ed2473ef5db7b42757f117707ba70b5273ed455a640aeea928e2cb2759a95
-size 395127
+oid sha256:03365e008789e6e85cbec250e337049bd55a712edd912daeaf92251675856e1b
+size 397245