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": 74523.163807717,
-      "required_pages_per_second_upper_estimate": 0.006459478961621039
+      "required_token_positions_per_second": 75442.08157597073,
+      "required_pages_per_second_upper_estimate": 0.006539128424797481
     },
     {
       "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": 9315.395475964626,
-      "required_pages_per_second_upper_estimate": 0.0037160505329362635
+      "required_token_positions_per_second": 9430.260196996342,
+      "required_pages_per_second_upper_estimate": 0.0037618717875364375
     },
     {
       "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": 18630.79095192925,
-      "required_pages_per_second_upper_estimate": 0.004309905596006937
+      "required_token_positions_per_second": 18860.520393992683,
+      "required_pages_per_second_upper_estimate": 0.00436304945932822
     },
     {
       "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": 27946.186427893877,
-      "required_pages_per_second_upper_estimate": 0.001701776131150119
+      "required_token_positions_per_second": 28290.780590989027,
+      "required_pages_per_second_upper_estimate": 0.0017227601077367653
     },
     {
       "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": 27946.186427893877,
-      "required_pages_per_second_upper_estimate": 0.0021513747912145632
+      "required_token_positions_per_second": 28290.780590989027,
+      "required_pages_per_second_upper_estimate": 0.0021779026037873233
     },
     {
       "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": 27946.186427893877,
-      "required_pages_per_second_upper_estimate": 0.34699810556506794
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28290.780590989027,
+      "required_pages_per_second_upper_estimate": 0.35127681178823433
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11276482227373208,
+    "producer_batch_seconds": 0.1209328845235973,
     "producer_p95_seconds": 0.35647802599851275,
-    "gpu_update_seconds": 0.9893299778607271,
-    "producer_batches_per_second": 8.868013799308264,
-    "gpu_batches_per_second": 1.0107850993885228,
-    "producer_headroom": 8.773391895738268,
+    "gpu_update_seconds": 0.9772795031610488,
+    "producer_batches_per_second": 8.269049431338694,
+    "gpu_batches_per_second": 1.0232487192921378,
+    "producer_headroom": 8.081172519872831,
     "sustainable": true,
-    "buffer_seconds": 60.34912864950435,
+    "buffer_seconds": 60.591329195985026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473529.6646580829,
-    "upload_bytes_per_second": 5598881.525655733,
+    "generation_bytes_per_second": 472432.1860072038,
+    "upload_bytes_per_second": 5627511.096575364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.60844282919754,
+    "estimated_full_shards_per_hour": 101.37294945871436,
     "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": 1.0177543880017765,
-      "bytes": 5824791,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.107241077999788,
       "bytes": 9747301,
@@ -562,6 +556,12 @@
       "bytes": 4940868,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 0.9655562699990696,
+      "bytes": 7648817,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 310300, "loss": 0.6191430687904358, "lr": 0.0003, "grad_norm": 0.14687646478434319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45419718011, "tokens_per_second": 151917.81851201394, "data_wait_s": 1.1657997674774379e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310400, "loss": 0.607383145019412, "lr": 0.0003, "grad_norm": 0.17157508109089079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45434509296, "tokens_per_second": 151587.65096832925, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310500, "loss": 0.579857038334012, "lr": 0.0003, "grad_norm": 0.13863951613268993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45449016212, "tokens_per_second": 148815.76590939896, "data_wait_s": 1.005100057227537e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 310600, "loss": 0.6004833858460188, "lr": 0.0003, "grad_norm": 0.14692711976080622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45463602971, "tokens_per_second": 149616.04517196465, "data_wait_s": 1.473700103815645e-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:a68151fbad0fac8696a82f5b22dfb68ed8937e9803388281764b3f817b1848d2
-size 646617
+oid sha256:cc3250b886963ed82daa839528fe3f1074ff3a3cd8efcd63e5f239261f985bd8
+size 648735