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": 74454.29039832999,
-      "required_pages_per_second_upper_estimate": 0.006453509189053433
+      "required_token_positions_per_second": 75627.32412712398,
+      "required_pages_per_second_upper_estimate": 0.006555184779638491
     },
     {
       "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": 9306.786299791249,
-      "required_pages_per_second_upper_estimate": 0.0037126162038420493
+      "required_token_positions_per_second": 9453.415515890498,
+      "required_pages_per_second_upper_estimate": 0.0037711087904461855
     },
     {
       "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": 18613.572599582498,
-      "required_pages_per_second_upper_estimate": 0.004305922433224167
+      "required_token_positions_per_second": 18906.831031780996,
+      "required_pages_per_second_upper_estimate": 0.0043737626103411585
     },
     {
       "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": 27920.35889937375,
-      "required_pages_per_second_upper_estimate": 0.001700203370169813
+      "required_token_positions_per_second": 28360.246547671497,
+      "required_pages_per_second_upper_estimate": 0.0017269902200390199
     },
     {
       "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": 27920.35889937375,
-      "required_pages_per_second_upper_estimate": 0.0021493865165739087
+      "required_token_positions_per_second": 28360.246547671497,
+      "required_pages_per_second_upper_estimate": 0.0021832502854268153
     },
     {
       "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": 27920.35889937375,
-      "required_pages_per_second_upper_estimate": 0.34667741410002545
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28360.246547671497,
+      "required_pages_per_second_upper_estimate": 0.35213934648262907
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 688946,
-    "producer_batch_seconds": 0.10249432328106423,
-    "producer_p95_seconds": 0.39124788300250657,
-    "gpu_update_seconds": 0.9902451504883822,
-    "producer_batches_per_second": 9.756637909182132,
-    "gpu_batches_per_second": 1.0098509439877659,
-    "producer_headroom": 9.661463374638716,
+    "producer_batch_seconds": 0.16837482361677303,
+    "producer_p95_seconds": 0.5921860819944413,
+    "gpu_update_seconds": 0.9748857420377407,
+    "producer_batches_per_second": 5.939130200819304,
+    "gpu_batches_per_second": 1.0257612321929794,
+    "producer_headroom": 5.789973352884482,
     "sustainable": true,
-    "buffer_seconds": 60.40495417979132,
+    "buffer_seconds": 60.44291600633992,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328429.3339946892,
-    "upload_bytes_per_second": 4846781.735776416,
+    "generation_bytes_per_second": 363332.71122008836,
+    "upload_bytes_per_second": 4833264.740068988,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.47328963165768,
+    "estimated_full_shards_per_hour": 77.96274187519062,
     "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": 1.207307631000731,
-      "bytes": 6036390,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.257034783000563,
-      "bytes": 2415905,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.2379260390007403,
       "bytes": 10087351,
@@ -562,6 +550,18 @@
       "bytes": 6232998,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9818889380039764,
+      "bytes": 963021,
+      "pages": 27,
+      "shards": 1
+    },
+    {
+      "seconds": 1.6556295810005395,
+      "bytes": 7143913,
+      "pages": 21,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -398,3 +398,4 @@
 {"step": 194800, "loss": 0.4885460287332535, "lr": 0.0003, "grad_norm": 0.15674144452471456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28510869924, "tokens_per_second": 150776.8410855801, "data_wait_s": 8.791997970547527e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194900, "loss": 0.6080953255295753, "lr": 0.0003, "grad_norm": 0.16244985294831388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28525593721, "tokens_per_second": 150274.05671662698, "data_wait_s": 1.1323994840495288e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195000, "loss": 0.587149441242218, "lr": 0.0003, "grad_norm": 0.14609337480068943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28540343020, "tokens_per_second": 150961.18382806727, "data_wait_s": 1.727200287859887e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 195100, "loss": 0.5759536698460579, "lr": 0.0003, "grad_norm": 0.1433815500195194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28555098636, "tokens_per_second": 135442.66935053776, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:28d8f8a650a7a511cf084b74a9cdc9cf3ccfee73a0f2defb79aab6e79c99b5d8
-size 853629
+oid sha256:de5c69881805e6b85a6877fcb4735740247b2b15ded614b964859a1399c8e8e8
+size 855747