Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75435.87408108677,
-      "required_pages_per_second_upper_estimate": 0.0065385903748736915
+      "required_token_positions_per_second": 75740.0202322792,
+      "required_pages_per_second_upper_estimate": 0.006564952992407678
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9429.484260135847,
-      "required_pages_per_second_upper_estimate": 0.0037615622547214965
+      "required_token_positions_per_second": 9467.5025290349,
+      "required_pages_per_second_upper_estimate": 0.0037767283106091036
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18858.968520271694,
-      "required_pages_per_second_upper_estimate": 0.004362690460654914
+      "required_token_positions_per_second": 18935.0050580698,
+      "required_pages_per_second_upper_estimate": 0.004380280175477146
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28288.452780407537,
-      "required_pages_per_second_upper_estimate": 0.0017226183562854308
+      "required_token_positions_per_second": 28402.5075871047,
+      "required_pages_per_second_upper_estimate": 0.0017295636956139642
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28288.452780407537,
-      "required_pages_per_second_upper_estimate": 0.002177723402484968
+      "required_token_positions_per_second": 28402.5075871047,
+      "required_pages_per_second_upper_estimate": 0.002186503657228428
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28288.452780407537,
-      "required_pages_per_second_upper_estimate": 0.35124790817149304
+      "required_token_positions_per_second": 28402.5075871047,
+      "required_pages_per_second_upper_estimate": 0.35266408715378894
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11273154383565043,
-    "producer_p95_seconds": 0.3810734330036212,
-    "gpu_update_seconds": 0.9773599218953709,
-    "producer_batches_per_second": 8.870631643773853,
-    "gpu_batches_per_second": 1.0231645247543235,
-    "producer_headroom": 8.66979985052142,
+    "producer_batch_seconds": 0.12112509722601317,
+    "producer_p95_seconds": 0.35556464300316293,
+    "gpu_update_seconds": 0.9734351769895394,
+    "producer_batches_per_second": 8.255927325565334,
+    "gpu_batches_per_second": 1.0272897709456272,
+    "producer_headroom": 8.036610077374466,
     "sustainable": true,
-    "buffer_seconds": 60.596315157513,
+    "buffer_seconds": 60.35298097335144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 523260.0847374552,
-    "upload_bytes_per_second": 5606227.952923502,
+    "generation_bytes_per_second": 521872.84342519596,
+    "upload_bytes_per_second": 5611208.983560376,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.27943331330053,
+    "estimated_full_shards_per_hour": 111.98176362101468,
     "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": 1.52958795199811,
-      "bytes": 10493875,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.2153502640030638,
       "bytes": 6178782,
@@ -562,6 +556,12 @@
       "bytes": 1148268,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 0.9499172950017964,
+      "bytes": 7609450,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -317,3 +317,4 @@
 {"step": 311900, "loss": 0.5738855935633183, "lr": 0.0003, "grad_norm": 0.14206522713526085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45653498694, "tokens_per_second": 149941.06140293326, "data_wait_s": 7.285998435690999e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 312000, "loss": 0.5994717217981815, "lr": 0.0003, "grad_norm": 0.16673448312736422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45668000901, "tokens_per_second": 148793.65612606192, "data_wait_s": 1.4966004528105259e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312100, "loss": 0.5786622334271669, "lr": 0.0003, "grad_norm": 0.15417092769651447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45682679327, "tokens_per_second": 150751.61326526327, "data_wait_s": 1.0034003935288638e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 312200, "loss": 0.5382980741560459, "lr": 0.0003, "grad_norm": 0.17255988027451766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45697309374, "tokens_per_second": 149458.23799435908, "data_wait_s": 1.0853000276256353e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ddab875f5f922ada2b84268b94ff88ea06cf1091d30351cb452bafbebd89613
-size 680505
+oid sha256:bc39af1362ae3ebb90239c761e4cb1c2eabbb8e53850561f8f7f8c7dab45e070
+size 682623