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": 74152.73405781625,
-      "required_pages_per_second_upper_estimate": 0.006427371049745259
+      "required_token_positions_per_second": 74957.95188930034,
+      "required_pages_per_second_upper_estimate": 0.006497165290572364
     },
     {
       "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": 9269.091757227032,
-      "required_pages_per_second_upper_estimate": 0.003697579287229548
+      "required_token_positions_per_second": 9369.743986162543,
+      "required_pages_per_second_upper_estimate": 0.003737730966236853
     },
     {
       "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": 18538.183514454064,
-      "required_pages_per_second_upper_estimate": 0.004288482495182286
+      "required_token_positions_per_second": 18739.487972325085,
+      "required_pages_per_second_upper_estimate": 0.004335050738673291
     },
     {
       "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": 27807.275271681094,
-      "required_pages_per_second_upper_estimate": 0.0016933171705472718
+      "required_token_positions_per_second": 28109.23195848763,
+      "required_pages_per_second_upper_estimate": 0.001711704748529491
     },
     {
       "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": 27807.275271681094,
-      "required_pages_per_second_upper_estimate": 0.0021406810258786115
+      "required_token_positions_per_second": 28109.23195848763,
+      "required_pages_per_second_upper_estimate": 0.002163926487498586
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27807.275271681094,
-      "required_pages_per_second_upper_estimate": 0.3452732939106385
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28109.23195848763,
+      "required_pages_per_second_upper_estimate": 0.34902258537675396
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09241123112498428,
-    "producer_p95_seconds": 0.21596320400021796,
-    "gpu_update_seconds": 0.994272172655359,
-    "producer_batches_per_second": 10.82119551732322,
-    "gpu_batches_per_second": 1.0057608243518914,
-    "producer_headroom": 10.759213577737391,
+    "producer_batch_seconds": 0.06487477017182641,
+    "producer_p95_seconds": 0.2090967020012613,
+    "gpu_update_seconds": 0.9835914421579078,
+    "producer_batches_per_second": 15.414312796044038,
+    "gpu_batches_per_second": 1.0166822901652066,
+    "producer_headroom": 15.16138615293405,
     "sustainable": true,
-    "buffer_seconds": 60.6506025319769,
+    "buffer_seconds": 60.98266941379028,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403273.35070770915,
+    "generation_bytes_per_second": 405600.07525056845,
     "upload_bytes_per_second": 5308221.154932217,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.5330733387323,
+    "estimated_full_shards_per_hour": 87.03233426225462,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -111,3 +111,4 @@
 {"step": 206100, "loss": 0.6121695507317781, "lr": 0.0003, "grad_norm": 0.14673832847579008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30165066372, "tokens_per_second": 147177.7628433172, "data_wait_s": 2.0351999410195276e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206200, "loss": 0.5746875181794167, "lr": 0.0003, "grad_norm": 0.14248435960817618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30179754206, "tokens_per_second": 147307.53043465337, "data_wait_s": 1.621200135559775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206300, "loss": 0.5879179909825325, "lr": 0.0003, "grad_norm": 0.13954754362615665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30194524871, "tokens_per_second": 149234.82555407833, "data_wait_s": 1.6312000298057683e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 206400, "loss": 0.6384740583598614, "lr": 0.0003, "grad_norm": 0.17451350463504176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30209255939, "tokens_per_second": 148678.38793833062, "data_wait_s": 2.5233000997104682e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a134a1275d39dd8cb54c170ada224c6a49ecad93f4130b973d88353b1a434feb
-size 241065
+oid sha256:f1c72f7777e37ed02b2c32c0530514a44e655cf1544efd6968c02346e8ba7032
+size 243183