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": 77028.49960927841,
-      "required_pages_per_second_upper_estimate": 0.005329010333713265
+      "required_token_positions_per_second": 77528.60805783985,
+      "required_pages_per_second_upper_estimate": 0.005363608996596219
     },
     {
       "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": 9628.562451159802,
-      "required_pages_per_second_upper_estimate": 0.00403127803105251
+      "required_token_positions_per_second": 9691.076007229982,
+      "required_pages_per_second_upper_estimate": 0.0040574511515475975
     },
     {
       "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": 19257.124902319603,
-      "required_pages_per_second_upper_estimate": 0.004010751272618434
+      "required_token_positions_per_second": 19382.152014459964,
+      "required_pages_per_second_upper_estimate": 0.0040367911228905975
     },
     {
       "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": 28885.687353479403,
-      "required_pages_per_second_upper_estimate": 0.0017316528509503154
+      "required_token_positions_per_second": 29073.228021689945,
+      "required_pages_per_second_upper_estimate": 0.0017428956276515177
     },
     {
       "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": 28885.687353479403,
-      "required_pages_per_second_upper_estimate": 0.0022457351370219116
+      "required_token_positions_per_second": 29073.228021689945,
+      "required_pages_per_second_upper_estimate": 0.0022603155990710603
     },
     {
       "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": 28885.687353479403,
-      "required_pages_per_second_upper_estimate": 0.3705145823357756
+      "required_token_positions_per_second": 29073.228021689945,
+      "required_pages_per_second_upper_estimate": 0.3729201526620996
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06160657856278817,
-    "producer_p95_seconds": 0.20875319200058584,
-    "gpu_update_seconds": 0.9571522277336316,
-    "producer_batches_per_second": 16.232032736257548,
-    "gpu_batches_per_second": 1.0447658909678603,
-    "producer_headroom": 15.536526294154147,
+    "producer_batch_seconds": 0.08143647875036208,
+    "producer_p95_seconds": 0.2212987160019111,
+    "gpu_update_seconds": 0.9509779918271661,
+    "producer_batches_per_second": 12.279509322418411,
+    "gpu_batches_per_second": 1.0515490459233923,
+    "producer_headroom": 11.677543116056425,
     "sustainable": true,
-    "buffer_seconds": 60.300590347218794,
+    "buffer_seconds": 60.86259147693863,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271865.9267407478,
-    "upload_bytes_per_second": 5196426.251204057,
+    "generation_bytes_per_second": 308493.9539983528,
+    "upload_bytes_per_second": 5212773.054563939,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.33609916369272,
+    "estimated_full_shards_per_hour": 66.19562115633906,
     "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": 1.1801391750013863,
-      "bytes": 2749619,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1686728980002954,
       "bytes": 6121581,
@@ -562,6 +556,12 @@
       "bytes": 1209380,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.5094498580001527,
+      "bytes": 5827665,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -166,3 +166,4 @@
 {"step": 901600, "loss": 0.4958744943141937, "lr": 3e-05, "grad_norm": 0.3201443174944292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131968518105, "tokens_per_second": 152118.64782818613, "data_wait_s": 8.743001671973616e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 901700, "loss": 0.5292997360229492, "lr": 3e-05, "grad_norm": 0.31630968411192373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131983219989, "tokens_per_second": 153078.70515092195, "data_wait_s": 1.0854000720428303e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 901800, "loss": 0.5543702840805054, "lr": 3e-05, "grad_norm": 0.321183228617444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131997800547, "tokens_per_second": 151369.25784464725, "data_wait_s": 8.856000931700692e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 901900, "loss": 0.5142814517021179, "lr": 3e-05, "grad_norm": 0.36111615552078913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132012334157, "tokens_per_second": 152130.04555850985, "data_wait_s": 8.2970000221394e-06, "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:fcd6b1734da2916e4663c02675d56917511a138561f3ce71ee614a454f9cb1ff
-size 358338
+oid sha256:6026ca6e9c250eebe2f8e7aa270a41557302bb0d491ad92dfee4548a96793e53
+size 360456