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": 73035.77562881671,
-      "required_pages_per_second_upper_estimate": 0.0063305559240248365
+      "required_token_positions_per_second": 75691.65435691421,
+      "required_pages_per_second_upper_estimate": 0.006560760760913231
     },
     {
       "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": 9129.47195360209,
-      "required_pages_per_second_upper_estimate": 0.0036418828600614686
+      "required_token_positions_per_second": 9461.456794614276,
+      "required_pages_per_second_upper_estimate": 0.0037743165767569315
     },
     {
       "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": 18258.94390720418,
-      "required_pages_per_second_upper_estimate": 0.004223885326494263
+      "required_token_positions_per_second": 18922.913589228552,
+      "required_pages_per_second_upper_estimate": 0.004377483026963307
     },
     {
       "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": 27388.415860806268,
-      "required_pages_per_second_upper_estimate": 0.0016678108300104853
+      "required_token_positions_per_second": 28384.370383842826,
+      "required_pages_per_second_upper_estimate": 0.0017284592350938672
     },
     {
       "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": 27388.415860806268,
-      "required_pages_per_second_upper_estimate": 0.0021084360689524084
+      "required_token_positions_per_second": 28384.370383842826,
+      "required_pages_per_second_upper_estimate": 0.002185107405114337
     },
     {
       "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": 27388.415860806268,
-      "required_pages_per_second_upper_estimate": 0.34007246185984413
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28384.370383842826,
+      "required_pages_per_second_upper_estimate": 0.35243888379059096
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13583734817956383,
-    "producer_p95_seconds": 0.40969298100390006,
-    "gpu_update_seconds": 1.0094778807402733,
-    "producer_batches_per_second": 7.361745597963947,
-    "gpu_batches_per_second": 0.9906111060766156,
-    "producer_headroom": 7.431519344781681,
+    "producer_batch_seconds": 0.2555313578981213,
+    "producer_p95_seconds": 0.39806883900018875,
+    "gpu_update_seconds": 0.9740571880269012,
+    "producer_batches_per_second": 3.913414025681707,
+    "gpu_batches_per_second": 1.0266337667767227,
+    "producer_headroom": 3.811889061440559,
     "sustainable": true,
-    "buffer_seconds": 60.5686728444164,
+    "buffer_seconds": 60.391545657667876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464088.0755223383,
-    "upload_bytes_per_second": 3818969.813836085,
+    "generation_bytes_per_second": 445202.18390778836,
+    "upload_bytes_per_second": 3734321.560257869,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.58249758961307,
+    "estimated_full_shards_per_hour": 95.53002489018668,
     "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": 1.2339288350012794,
-      "bytes": 10876745,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0449587049988622,
       "bytes": 2671135,
@@ -562,6 +556,12 @@
       "bytes": 6390118,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1548123519969522,
+      "bytes": 1183539,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -368,3 +368,4 @@
 {"step": 351800, "loss": 0.5804985966533422, "lr": 0.0003, "grad_norm": 0.17546815883735956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51494197641, "tokens_per_second": 148130.76413141593, "data_wait_s": 1.1023999832104892e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351900, "loss": 0.5837832074612379, "lr": 0.0003, "grad_norm": 0.15345544160138913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51508829587, "tokens_per_second": 148419.20414938993, "data_wait_s": 1.3760000001639128e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352000, "loss": 0.6193990055471659, "lr": 0.0003, "grad_norm": 0.16301558693260768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51523348305, "tokens_per_second": 148386.81821898517, "data_wait_s": 1.5316996723413467e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 352100, "loss": 0.5566759584471583, "lr": 0.0003, "grad_norm": 0.1369592259047149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51538001908, "tokens_per_second": 148347.85974753468, "data_wait_s": 1.5137004083953798e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:403daf3a15e48bb52e55c30c6e14a6e5e9ba2e64db66cae736a1719af78e436a
-size 789306
+oid sha256:9acf247668d51102c18d9492d221eefefa1571dba3ae6857019f294a25473346
+size 791424