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": 75474.8105704073,
-      "required_pages_per_second_upper_estimate": 0.005221522520948911
+      "required_token_positions_per_second": 74922.43979939033,
+      "required_pages_per_second_upper_estimate": 0.005183308229333188
     },
     {
       "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": 9434.351321300912,
-      "required_pages_per_second_upper_estimate": 0.003949965886570161
+      "required_token_positions_per_second": 9365.30497492379,
+      "required_pages_per_second_upper_estimate": 0.0039210576231937305
     },
     {
       "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": 18868.702642601824,
-      "required_pages_per_second_upper_estimate": 0.003929853159303046
+      "required_token_positions_per_second": 18730.60994984758,
+      "required_pages_per_second_upper_estimate": 0.0039010920931515426
     },
     {
       "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": 28303.053963902737,
-      "required_pages_per_second_upper_estimate": 0.001696724868874869
+      "required_token_positions_per_second": 28095.914924771372,
+      "required_pages_per_second_upper_estimate": 0.001684307199761943
     },
     {
       "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": 28303.053963902737,
-      "required_pages_per_second_upper_estimate": 0.0022004379537157682
+      "required_token_positions_per_second": 28095.914924771372,
+      "required_pages_per_second_upper_estimate": 0.002184333804531649
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28303.053963902737,
-      "required_pages_per_second_upper_estimate": 0.36304118679727987
+      "required_token_positions_per_second": 28095.914924771372,
+      "required_pages_per_second_upper_estimate": 0.36038422961187483
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06453229392991489,
-    "producer_p95_seconds": 0.2151350989988714,
-    "gpu_update_seconds": 0.9768557144137808,
-    "producer_batches_per_second": 15.496117356157324,
-    "gpu_batches_per_second": 1.0236926346897692,
-    "producer_headroom": 15.137470790588852,
+    "producer_batch_seconds": 0.09445721792189943,
+    "producer_p95_seconds": 0.2508726790001674,
+    "gpu_update_seconds": 0.9840576494493704,
+    "producer_batches_per_second": 10.586803443934114,
+    "gpu_batches_per_second": 1.0162006266193349,
+    "producer_headroom": 10.418024912220304,
     "sustainable": true,
-    "buffer_seconds": 60.56505429365441,
+    "buffer_seconds": 60.02751661641159,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277186.4024307882,
-    "upload_bytes_per_second": 4943741.715317381,
+    "generation_bytes_per_second": 276098.538215482,
+    "upload_bytes_per_second": 4894283.470154734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.477749392440174,
+    "estimated_full_shards_per_hour": 59.24431905601831,
     "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": 1.2842068059999292,
-      "bytes": 7447718,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.3457034299990482,
       "bytes": 1181953,
@@ -562,6 +556,12 @@
       "bytes": 11006649,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8713928039978782,
+      "bytes": 1169048,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -167,3 +167,4 @@
 {"step": 736700, "loss": 0.5290622115135193, "lr": 5e-05, "grad_norm": 0.2205401782882761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107832514860, "tokens_per_second": 151377.62170462927, "data_wait_s": 1.65250021382235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736800, "loss": 0.5798033475875854, "lr": 5e-05, "grad_norm": 0.22902320557757003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107847215707, "tokens_per_second": 150690.55532955113, "data_wait_s": 1.414399957866408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736900, "loss": 0.49935975670814514, "lr": 5e-05, "grad_norm": 0.2256634917099695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107861781751, "tokens_per_second": 148566.06792430306, "data_wait_s": 1.6441998013760895e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 737000, "loss": 0.5023998022079468, "lr": 5e-05, "grad_norm": 0.22182048485091402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107876384978, "tokens_per_second": 149998.93509704448, "data_wait_s": 1.329600127064623e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83bf544be077052680162951c0920c1da19d905be9340397fbdc500281bed311
-size 360456
+oid sha256:8c20fe0aa9d4b0a67dc15f22e108c6385c90f7ac5fb33e22bba0f827ac105ccc
+size 362574