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": 76306.56527147602,
-      "required_pages_per_second_upper_estimate": 0.0066140596805099846
+      "required_token_positions_per_second": 75619.2635077979,
+      "required_pages_per_second_upper_estimate": 0.00655448610558476
     },
     {
       "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": 9538.320658934503,
-      "required_pages_per_second_upper_estimate": 0.003804978721451453
+      "required_token_positions_per_second": 9452.407938474738,
+      "required_pages_per_second_upper_estimate": 0.003770706852750414
     },
     {
       "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": 19076.641317869005,
-      "required_pages_per_second_upper_estimate": 0.004413045231468657
+      "required_token_positions_per_second": 18904.815876949477,
+      "required_pages_per_second_upper_estimate": 0.0043732964397364005
     },
     {
       "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": 28614.961976803505,
-      "required_pages_per_second_upper_estimate": 0.001742501053284584
+      "required_token_positions_per_second": 28357.22381542422,
+      "required_pages_per_second_upper_estimate": 0.001726806151504211
     },
     {
       "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": 28614.961976803505,
-      "required_pages_per_second_upper_estimate": 0.0022028589842588335
+      "required_token_positions_per_second": 28357.22381542422,
+      "required_pages_per_second_upper_estimate": 0.0021830175871309612
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28614.961976803505,
-      "required_pages_per_second_upper_estimate": 0.3553020596347456
+      "required_token_positions_per_second": 28357.22381542422,
+      "required_pages_per_second_upper_estimate": 0.35210181426455195
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1089492368281526,
-    "producer_p95_seconds": 0.19254466900019906,
-    "gpu_update_seconds": 0.9662078189169929,
-    "producer_batches_per_second": 9.178586551985822,
-    "gpu_batches_per_second": 1.0349740298323027,
-    "producer_headroom": 8.868422093135063,
+    "producer_batch_seconds": 0.10616335296094803,
+    "producer_p95_seconds": 0.2060124370000267,
+    "gpu_update_seconds": 0.9749896597762701,
+    "producer_batches_per_second": 9.419446278866568,
+    "gpu_batches_per_second": 1.025651903046304,
+    "producer_headroom": 9.183862722712968,
     "sustainable": true,
-    "buffer_seconds": 60.871092591770555,
+    "buffer_seconds": 60.449358906128744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347446.2962334346,
-    "upload_bytes_per_second": 5599964.374492737,
+    "generation_bytes_per_second": 354813.2692556205,
+    "upload_bytes_per_second": 5591021.14151757,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.55388703586844,
+    "estimated_full_shards_per_hour": 76.13466795207464,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 7370564,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0604308100009803,
+      "bytes": 5273416,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -75,3 +75,4 @@
 {"step": 287700, "loss": 0.5897219739854336, "lr": 0.0003, "grad_norm": 0.15187574616742733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42111467830, "tokens_per_second": 150928.29858113185, "data_wait_s": 1.4171000657370314e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287800, "loss": 0.5792989619076252, "lr": 0.0003, "grad_norm": 0.15761427272184975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42125903408, "tokens_per_second": 148096.6619555482, "data_wait_s": 9.186000170302577e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 287900, "loss": 0.5967279635369778, "lr": 0.0003, "grad_norm": 0.1422105488783368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42140400954, "tokens_per_second": 148699.04759279452, "data_wait_s": 1.422199966327753e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 288000, "loss": 0.5911591537296772, "lr": 0.0003, "grad_norm": 0.14532814156487564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42154937808, "tokens_per_second": 149293.68317400984, "data_wait_s": 1.0351999662816525e-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:d17ffee25ab6dcc618f3d1bdb9fc5ae0702b3a156386e7594de5202da6535946
-size 164034
+oid sha256:10147700e0189f62eca0bf6351d15f36a13c2a1d0f9cab19265e7552643683d5
+size 166152