Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77684.12266091979,
-      "required_pages_per_second_upper_estimate": 0.0053743678576808835
+      "required_token_positions_per_second": 75624.18374599694,
+      "required_pages_per_second_upper_estimate": 0.005231856503829722
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9710.515332614974,
-      "required_pages_per_second_upper_estimate": 0.004065589991147019
+      "required_token_positions_per_second": 9453.022968249617,
+      "required_pages_per_second_upper_estimate": 0.003957783315239257
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19421.030665229948,
-      "required_pages_per_second_upper_estimate": 0.004044888520547016
+      "required_token_positions_per_second": 18906.045936499235,
+      "required_pages_per_second_upper_estimate": 0.003937630782612992
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29131.54599784492,
-      "required_pages_per_second_upper_estimate": 0.001746391701275615
+      "required_token_positions_per_second": 28359.068904748856,
+      "required_pages_per_second_upper_estimate": 0.0017000828790487315
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29131.54599784492,
-      "required_pages_per_second_upper_estimate": 0.0022648495652034415
+      "required_token_positions_per_second": 28359.068904748856,
+      "required_pages_per_second_upper_estimate": 0.002204792868982868
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29131.54599784492,
-      "required_pages_per_second_upper_estimate": 0.3736681930432514
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28359.068904748856,
+      "required_pages_per_second_upper_estimate": 0.36375968631429634
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2787840826093202,
-    "producer_p95_seconds": 0.2523256790009327,
-    "gpu_update_seconds": 0.9490742441903127,
-    "producer_batches_per_second": 3.587005365013506,
-    "gpu_batches_per_second": 1.053658347722979,
-    "producer_headroom": 3.40433440570679,
+    "producer_batch_seconds": 0.2389485830469198,
+    "producer_p95_seconds": 0.266439843999251,
+    "gpu_update_seconds": 0.9749262252883845,
+    "producer_batches_per_second": 4.1850007530851965,
+    "gpu_batches_per_second": 1.0257186380479186,
+    "producer_headroom": 4.080066987034397,
     "sustainable": true,
-    "buffer_seconds": 60.740751628180014,
+    "buffer_seconds": 60.445425967879835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 34909.72651549879,
-    "upload_bytes_per_second": 5190724.488005381,
+    "generation_bytes_per_second": 32844.7144454393,
+    "upload_bytes_per_second": 5014224.279017709,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 7.490814653384426,
+    "estimated_full_shards_per_hour": 7.047711134170382,
     "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": 1.245589108999411,
-      "bytes": 10344850,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3122096120023343,
-      "bytes": 13394029,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.2143651210026292,
       "bytes": 938191,
@@ -562,6 +550,18 @@
       "bytes": 5955841,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.0264875259999826,
+      "bytes": 2166910,
+      "pages": 21,
+      "shards": 2
+    },
+    {
+      "seconds": 1.2000135250018502,
+      "bytes": 5803444,
+      "pages": 1,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 710700, "loss": 0.5578862428665161, "lr": 8e-05, "grad_norm": 0.20971052914242538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104026390979, "tokens_per_second": 153050.39375180213, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710800, "loss": 0.5703843832015991, "lr": 8e-05, "grad_norm": 0.215808214121722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104041158263, "tokens_per_second": 154485.9335840067, "data_wait_s": 1.791400063666515e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710900, "loss": 0.5743618011474609, "lr": 8e-05, "grad_norm": 0.21989600077509383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104055896511, "tokens_per_second": 43432.19885658717, "data_wait_s": 1.4179997378960252e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 711000, "loss": 0.5434914827346802, "lr": 8e-05, "grad_norm": 0.23269289903401955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104070684528, "tokens_per_second": 154419.78689260912, "data_wait_s": 1.8021000869339332e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96196ca78efae975d9d2e6b602b1ae7f221a301457d45ace8edcc3e8feb152ce
-size 552642
+oid sha256:b7d2926d2a6d4e5bef0af082a0595a5bd786a50068d70aae82d094bb0cf5337d
+size 554760