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": 74747.26699173397,
-      "required_pages_per_second_upper_estimate": 0.006478903657627302
+      "required_token_positions_per_second": 75637.69536132636,
+      "required_pages_per_second_upper_estimate": 0.006556083731933501
     },
     {
       "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": 9343.408373966746,
-      "required_pages_per_second_upper_estimate": 0.003727225296779458
+      "required_token_positions_per_second": 9454.711920165795,
+      "required_pages_per_second_upper_estimate": 0.0037716259454945726
     },
     {
       "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": 18686.81674793349,
-      "required_pages_per_second_upper_estimate": 0.004322866177892185
+      "required_token_positions_per_second": 18909.42384033159,
+      "required_pages_per_second_upper_estimate": 0.004374362410967467
     },
     {
       "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": 28030.225121900236,
-      "required_pages_per_second_upper_estimate": 0.0017068936466981546
+      "required_token_positions_per_second": 28364.135760497382,
+      "required_pages_per_second_upper_estimate": 0.0017272270526949936
     },
     {
       "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": 28030.225121900236,
-      "required_pages_per_second_upper_estimate": 0.0021578443225131667
+      "required_token_positions_per_second": 28364.135760497382,
+      "required_pages_per_second_upper_estimate": 0.002183549687796182
     },
     {
       "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": 28030.225121900236,
-      "required_pages_per_second_upper_estimate": 0.348041584885211
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28364.135760497382,
+      "required_pages_per_second_upper_estimate": 0.35218763748956855
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10197708599213229,
-    "producer_p95_seconds": 0.3930859789979877,
-    "gpu_update_seconds": 0.9863638226151241,
-    "producer_batches_per_second": 9.80612448640817,
-    "gpu_batches_per_second": 1.0138246933557666,
-    "producer_headroom": 9.672406433453332,
+    "producer_batch_seconds": 0.1643137062970368,
+    "producer_p95_seconds": 0.41241809699567966,
+    "gpu_update_seconds": 0.9747520683674508,
+    "producer_batches_per_second": 6.085919565299428,
+    "gpu_batches_per_second": 1.0259019010596564,
+    "producer_headroom": 5.932262684193554,
     "sustainable": true,
-    "buffer_seconds": 60.16819317952257,
+    "buffer_seconds": 60.43462823878195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265135.997709445,
-    "upload_bytes_per_second": 5958138.732714116,
+    "generation_bytes_per_second": 254569.13765190306,
+    "upload_bytes_per_second": 5973007.604793909,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.89201305830491,
+    "estimated_full_shards_per_hour": 54.62461087386912,
     "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": 1.117633730002126,
-      "bytes": 7565346,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8082868439996673,
       "bytes": 1195987,
@@ -562,6 +556,12 @@
       "bytes": 5682376,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9258055570026045,
+      "bytes": 7457751,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -370,3 +370,4 @@
 {"step": 152000, "loss": 0.6232686080038548, "lr": 0.0003, "grad_norm": 0.1419686044015527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22245102282, "tokens_per_second": 149083.7484558147, "data_wait_s": 9.306000720243901e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 152100, "loss": 0.5953622683882713, "lr": 0.0003, "grad_norm": 0.13832643536771577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22259785078, "tokens_per_second": 149859.25446274687, "data_wait_s": 1.4823999663349241e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152200, "loss": 0.567689910531044, "lr": 0.0003, "grad_norm": 0.13095856199582015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22274231802, "tokens_per_second": 146985.4576452147, "data_wait_s": 1.0063995432574302e-05, "gpu_peak_gb": 85.46427917480469}
+{"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}

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 sha256:207885e24d44a4e10f25558eb7e9b9ccd464b91434d67dcc9825c3a1bb387eed
-size 793542
+oid sha256:d6641cf3d868395158b9ffea4fba9b82e5804677ec2bb2b24110a8e874589551
+size 795660