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": 74283.00610230482,
-      "required_pages_per_second_upper_estimate": 0.00643866270039005
+      "required_token_positions_per_second": 73231.98518460449,
+      "required_pages_per_second_upper_estimate": 0.006347562871031948
     },
     {
       "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": 9285.375762788102,
-      "required_pages_per_second_upper_estimate": 0.003704075220515439
+      "required_token_positions_per_second": 9153.998148075561,
+      "required_pages_per_second_upper_estimate": 0.003651666725736222
     },
     {
       "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": 18570.751525576205,
-      "required_pages_per_second_upper_estimate": 0.004296016531377974
+      "required_token_positions_per_second": 18307.996296151123,
+      "required_pages_per_second_upper_estimate": 0.004235232733384579
     },
     {
       "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": 27856.127288364307,
-      "required_pages_per_second_upper_estimate": 0.0016962920020565564
+      "required_token_positions_per_second": 27461.994444226682,
+      "required_pages_per_second_upper_estimate": 0.001672291379703796
     },
     {
       "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": 27856.127288364307,
-      "required_pages_per_second_upper_estimate": 0.0021444417893539225
+      "required_token_positions_per_second": 27461.994444226682,
+      "required_pages_per_second_upper_estimate": 0.002114100351982667
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27856.127288364307,
-      "required_pages_per_second_upper_estimate": 0.34587987246066165
+      "required_token_positions_per_second": 27461.994444226682,
+      "required_pages_per_second_upper_estimate": 0.3409860616142479
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2301900081488384,
-    "producer_p95_seconds": 0.3881229530015844,
-    "gpu_update_seconds": 0.9925284916237713,
-    "producer_batches_per_second": 4.344237215341731,
-    "gpu_batches_per_second": 1.0075277520386396,
-    "producer_headroom": 4.3117792105989805,
+    "producer_batch_seconds": 0.14769628186007822,
+    "producer_p95_seconds": 0.3928780080022989,
+    "gpu_update_seconds": 1.006773199089785,
+    "producer_batches_per_second": 6.7706511457570855,
+    "gpu_batches_per_second": 0.9932723684977822,
+    "producer_headroom": 6.816510113934779,
     "sustainable": true,
-    "buffer_seconds": 60.54423798905005,
+    "buffer_seconds": 60.4063919453871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463175.29676784197,
-    "upload_bytes_per_second": 3803219.38662073,
+    "generation_bytes_per_second": 461466.9836228392,
+    "upload_bytes_per_second": 3785887.0474941544,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.38663651730008,
+    "estimated_full_shards_per_hour": 99.0200722838772,
     "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": 0.9684943469983409,
-      "bytes": 6504817,
-      "pages": 23,
-      "shards": 2
-    },
     {
       "seconds": 1.0333191259996966,
       "bytes": 7473380,
@@ -562,6 +556,12 @@
       "bytes": 7846458,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9881938749967958,
+      "bytes": 4661656,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -365,3 +365,4 @@
 {"step": 351500, "loss": 0.5582461468875408, "lr": 0.0003, "grad_norm": 0.14971862829255003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51450569203, "tokens_per_second": 147946.80826148984, "data_wait_s": 1.2718999641947448e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351600, "loss": 0.5729933995753527, "lr": 0.0003, "grad_norm": 0.1799935482318668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51465136218, "tokens_per_second": 147513.37203497015, "data_wait_s": 1.3928001862950623e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351700, "loss": 0.6278344821184874, "lr": 0.0003, "grad_norm": 0.1583050349370463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51479621936, "tokens_per_second": 146789.8006376949, "data_wait_s": 1.997900108108297e-05, "gpu_peak_gb": 85.46427917480469}
+{"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}

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:bfdbce68b22e269bb3b8cb97acef609bf3fa906ad916dca2928a03cab341d88c
-size 782952
+oid sha256:ef65a6465e2dea24c2a851842ec2479df9c6352689bfac8b8bea5e1642de8a17
+size 785070