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": 74991.77555115592,
-      "required_pages_per_second_upper_estimate": 0.006500097039856719
+      "required_token_positions_per_second": 73370.8022646075,
+      "required_pages_per_second_upper_estimate": 0.006359595183697936
     },
     {
       "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": 9373.97194389449,
-      "required_pages_per_second_upper_estimate": 0.003739417561789728
+      "required_token_positions_per_second": 9171.350283075937,
+      "required_pages_per_second_upper_estimate": 0.0036585887518254096
     },
     {
       "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": 18747.94388778898,
-      "required_pages_per_second_upper_estimate": 0.004337006865896834
+      "required_token_positions_per_second": 18342.700566151874,
+      "required_pages_per_second_upper_estimate": 0.004243260955474961
     },
     {
       "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": 28121.91583168347,
-      "required_pages_per_second_upper_estimate": 0.001712477129859978
+      "required_token_positions_per_second": 27514.05084922781,
+      "required_pages_per_second_upper_estimate": 0.0016754613416495152
     },
     {
       "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": 28121.91583168347,
-      "required_pages_per_second_upper_estimate": 0.0021649029271684096
+      "required_token_positions_per_second": 27514.05084922781,
+      "required_pages_per_second_upper_estimate": 0.0021181077981410053
     },
     {
       "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": 28121.91583168347,
-      "required_pages_per_second_upper_estimate": 0.3491800766316534
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27514.05084922781,
+      "required_pages_per_second_upper_estimate": 0.3416324279427817
     }
   ],
   "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": 0.1033568705856851,
-    "producer_p95_seconds": 0.33397296899784124,
-    "gpu_update_seconds": 0.9831478113184048,
-    "producer_batches_per_second": 9.675215535584334,
-    "gpu_batches_per_second": 1.0171410529399403,
-    "producer_headroom": 9.512166977843565,
+    "producer_batch_seconds": 0.13149880374209033,
+    "producer_p95_seconds": 0.36008762800338445,
+    "gpu_update_seconds": 1.0048683907544624,
+    "producer_batches_per_second": 7.604631917118486,
+    "gpu_batches_per_second": 0.9951551956462605,
+    "producer_headroom": 7.641654236834876,
     "sustainable": true,
-    "buffer_seconds": 60.9551643017411,
+    "buffer_seconds": 60.29210344526775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279358.5621608768,
-    "upload_bytes_per_second": 5937412.000890247,
+    "generation_bytes_per_second": 294275.83574370894,
+    "upload_bytes_per_second": 5973278.84412117,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.9438443052266,
+    "estimated_full_shards_per_hour": 63.144743959745895,
     "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": 0.9750485680015117,
-      "bytes": 2678604,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9471035349997692,
       "bytes": 5896771,
@@ -562,6 +556,12 @@
       "bytes": 6159261,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4083330920038861,
+      "bytes": 7782602,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 112000, "loss": 0.5638746749609709, "lr": 0.0003, "grad_norm": 0.1325248474738872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16388938660, "tokens_per_second": 148220.7544071908, "data_wait_s": 1.442500069970265e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112100, "loss": 0.5938390232622623, "lr": 0.0003, "grad_norm": 0.1428568940247015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16403572042, "tokens_per_second": 149137.73920208952, "data_wait_s": 1.1523996363393962e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112200, "loss": 0.5894890539348125, "lr": 0.0003, "grad_norm": 0.13322304199477333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16418162723, "tokens_per_second": 147506.0320255269, "data_wait_s": 9.723997209221125e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 112300, "loss": 0.5872876457870007, "lr": 0.0003, "grad_norm": 0.14929047942423196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16432892142, "tokens_per_second": 149901.6128204821, "data_wait_s": 1.7496997315902263e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa4bd06536306e6bf37c0a147abbe42af47c8c2646d82a44dec892338f318e2e
-size 580176
+oid sha256:71f0620b50be812f15f1160a814b81e2f420d4444c27ccba635bb4e2c28218bf
+size 582294