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": 75510.85034947233,
-      "required_pages_per_second_upper_estimate": 0.006545089127791736
+      "required_token_positions_per_second": 74562.52282175042,
+      "required_pages_per_second_upper_estimate": 0.006462890501202946
     },
     {
       "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": 9438.85629368404,
-      "required_pages_per_second_upper_estimate": 0.0037653008990282595
+      "required_token_positions_per_second": 9320.315352718802,
+      "required_pages_per_second_upper_estimate": 0.0037180131453322172
     },
     {
       "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": 18877.71258736808,
-      "required_pages_per_second_upper_estimate": 0.00436702657069335
+      "required_token_positions_per_second": 18640.630705437605,
+      "required_pages_per_second_upper_estimate": 0.004312181849807348
     },
     {
       "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": 28316.568881052124,
-      "required_pages_per_second_upper_estimate": 0.0017243304793009065
+      "required_token_positions_per_second": 27960.946058156405,
+      "required_pages_per_second_upper_estimate": 0.0017026749151953116
     },
     {
       "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": 28316.568881052124,
-      "required_pages_per_second_upper_estimate": 0.0021798878577429363
+      "required_token_positions_per_second": 27960.946058156405,
+      "required_pages_per_second_upper_estimate": 0.0021525110284094442
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28316.568881052124,
-      "required_pages_per_second_upper_estimate": 0.3515970160429632
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27960.946058156405,
+      "required_pages_per_second_upper_estimate": 0.34718137077562367
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.8705792851252454,
-    "producer_p95_seconds": 0.4052391040022485,
-    "gpu_update_seconds": 0.9763894812305635,
-    "producer_batches_per_second": 1.1486604575666368,
-    "gpu_batches_per_second": 1.024181455477869,
-    "producer_headroom": 1.1215399882735502,
+    "producer_batch_seconds": 0.1361986398049453,
+    "producer_p95_seconds": 0.3901895420058281,
+    "gpu_update_seconds": 0.9888077442907152,
+    "producer_batches_per_second": 7.342217230892571,
+    "gpu_batches_per_second": 1.0113189401821616,
+    "producer_headroom": 7.260041258171303,
     "sustainable": true,
-    "buffer_seconds": 60.53614783629494,
+    "buffer_seconds": 60.317272401733625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 237292.03330517287,
-    "upload_bytes_per_second": 5896013.988767982,
+    "generation_bytes_per_second": 238196.23220897221,
+    "upload_bytes_per_second": 5974153.714860875,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.91734647146596,
+    "estimated_full_shards_per_hour": 51.11136650754809,
     "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": 0.8454459890017461,
-      "bytes": 981466,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.0536112129993853,
       "bytes": 11943546,
@@ -562,6 +556,12 @@
       "bytes": 9878612,
       "pages": 14,
       "shards": 3
+    },
+    {
+      "seconds": 1.218867665003927,
+      "bytes": 8704871,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 152300, "loss": 0.61142473295331, "lr": 0.0003, "grad_norm": 0.13135204256711755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22288766647, "tokens_per_second": 148272.28295995138, "data_wait_s": 1.275399699807167e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152400, "loss": 0.5488371104001999, "lr": 0.0003, "grad_norm": 0.13606638197257243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22303225480, "tokens_per_second": 96123.30438182088, "data_wait_s": 1.4270997780840844e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152500, "loss": 0.6430874671787024, "lr": 0.0003, "grad_norm": 0.15707740170207865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22317954706, "tokens_per_second": 149912.69458676706, "data_wait_s": 1.2789001630153507e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 152600, "loss": 0.628054790198803, "lr": 0.0003, "grad_norm": 0.1676492831230582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22332668751, "tokens_per_second": 151222.65423052188, "data_wait_s": 1.2320000678300858e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b495b04edcdc88f58813fb888e82ce56005c0fa2619a3f00c5139f64bf6d513a
-size 799896
+oid sha256:727324d2d5c710307157a8891b0ecff2c7040f180f88dc132d8af323d08fcd86
+size 802014