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": 74992.96369194628,
-      "required_pages_per_second_upper_estimate": 0.006500200024889112
+      "required_token_positions_per_second": 73097.51014024274,
+      "required_pages_per_second_upper_estimate": 0.006335906915010556
     },
     {
       "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": 9374.120461493285,
-      "required_pages_per_second_upper_estimate": 0.0037394768076804232
+      "required_token_positions_per_second": 9137.188767530342,
+      "required_pages_per_second_upper_estimate": 0.0036449612125140987
     },
     {
       "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": 18748.24092298657,
-      "required_pages_per_second_upper_estimate": 0.004337075579762156
+      "required_token_positions_per_second": 18274.377535060685,
+      "required_pages_per_second_upper_estimate": 0.004227455624676293
     },
     {
       "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": 28122.361384479856,
-      "required_pages_per_second_upper_estimate": 0.001712504261687643
+      "required_token_positions_per_second": 27411.56630259103,
+      "required_pages_per_second_upper_estimate": 0.0016692205704541962
     },
     {
       "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": 28122.361384479856,
-      "required_pages_per_second_upper_estimate": 0.0021649372270560446
+      "required_token_positions_per_second": 27411.56630259103,
+      "required_pages_per_second_upper_estimate": 0.002110218254045524
     },
     {
       "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": 28122.361384479856,
-      "required_pages_per_second_upper_estimate": 0.34918560890621525
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27411.56630259103,
+      "required_pages_per_second_upper_estimate": 0.34035991286726636
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.337569198023175,
-    "producer_p95_seconds": 0.27750196600027266,
-    "gpu_update_seconds": 0.9831322349501687,
-    "producer_batches_per_second": 2.962355587701895,
-    "gpu_batches_per_second": 1.0171571681307818,
-    "producer_headroom": 2.912387269654485,
+    "producer_batch_seconds": 0.3984023113280557,
+    "producer_p95_seconds": 0.3022748049988877,
+    "gpu_update_seconds": 1.0086253260685298,
+    "producer_batches_per_second": 2.510025598663186,
+    "gpu_batches_per_second": 0.9914484339768167,
+    "producer_headroom": 2.5316753878920126,
     "sustainable": true,
-    "buffer_seconds": 60.954198566910456,
+    "buffer_seconds": 60.517519564111794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 183050.74172746114,
-    "upload_bytes_per_second": 6128555.531456836,
+    "generation_bytes_per_second": 177655.88366059362,
+    "upload_bytes_per_second": 6160295.317555943,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 39.278427971533546,
+    "estimated_full_shards_per_hour": 38.120817016252104,
     "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": 0.9855451060011546,
-      "bytes": 8735177,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.964613582000311,
       "bytes": 5800981,
@@ -562,6 +556,12 @@
       "bytes": 8209294,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 1.0802435129990045,
+      "bytes": 11583568,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -178,3 +178,4 @@
 {"step": 17800, "loss": 0.6375476289540529, "lr": 0.0003, "grad_norm": 0.1435686926483544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2604557371, "tokens_per_second": 142427.46112601698, "data_wait_s": 1.2294000043766573e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17900, "loss": 0.7036605700850487, "lr": 0.0003, "grad_norm": 0.1436700177685992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2619072920, "tokens_per_second": 148300.02777097083, "data_wait_s": 1.3864999345969409e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18000, "loss": 0.6562089202925563, "lr": 0.0003, "grad_norm": 0.13488393327309636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2633628500, "tokens_per_second": 148094.5539179273, "data_wait_s": 9.250998118659481e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 18100, "loss": 0.7005441393703222, "lr": 0.0003, "grad_norm": 0.14222939985207977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2648334775, "tokens_per_second": 149106.81930044683, "data_wait_s": 9.606999810785055e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b87d5a53e89390065b74a9266e608b3442965ae9f3dd2e3f12ca4eeaa220a196
-size 377980
+oid sha256:339a958f59e55488fd65f626207b542f1cbd8e3d5c08ee8793fb33e41bf6b98b
+size 380098