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": 75967.31794180794,
-      "required_pages_per_second_upper_estimate": 0.005255595323676953
+      "required_token_positions_per_second": 76461.51968750442,
+      "required_pages_per_second_upper_estimate": 0.005289785347150271
     },
     {
       "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": 9495.914742725992,
-      "required_pages_per_second_upper_estimate": 0.003975741205530413
+      "required_token_positions_per_second": 9557.689960938053,
+      "required_pages_per_second_upper_estimate": 0.004001605199382554
     },
     {
       "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": 18991.829485451985,
-      "required_pages_per_second_upper_estimate": 0.003955497233595532
+      "required_token_positions_per_second": 19115.379921876105,
+      "required_pages_per_second_upper_estimate": 0.003981229531258561
     },
     {
       "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": 28487.744228177973,
-      "required_pages_per_second_upper_estimate": 0.0017077967682124645
+      "required_token_positions_per_second": 28673.069882814158,
+      "required_pages_per_second_upper_estimate": 0.001718906758232014
     },
     {
       "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": 28487.744228177973,
-      "required_pages_per_second_upper_estimate": 0.002214796809396549
+      "required_token_positions_per_second": 28673.069882814158,
+      "required_pages_per_second_upper_estimate": 0.0022292050638831064
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28487.744228177973,
-      "required_pages_per_second_upper_estimate": 0.36541019520244705
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28673.069882814158,
+      "required_pages_per_second_upper_estimate": 0.3677873537129354
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.24069697325774086,
-    "producer_p95_seconds": 0.3421043490016018,
-    "gpu_update_seconds": 0.9705226141651693,
-    "producer_batches_per_second": 4.1546014744821465,
-    "gpu_batches_per_second": 1.03037269343815,
-    "producer_headroom": 4.03213468382888,
+    "producer_batch_seconds": 0.16419922164885747,
+    "producer_p95_seconds": 0.32828900899767177,
+    "gpu_update_seconds": 0.9642497337395827,
+    "producer_batches_per_second": 6.090162851919695,
+    "gpu_batches_per_second": 1.0370757336087297,
+    "producer_headroom": 5.8724379083942635,
     "sustainable": true,
-    "buffer_seconds": 60.172402078240495,
+    "buffer_seconds": 60.747733225593706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369371.7090103406,
-    "upload_bytes_per_second": 4604400.782809922,
+    "generation_bytes_per_second": 354963.9155388895,
+    "upload_bytes_per_second": 4630086.324175555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.2585702203051,
+    "estimated_full_shards_per_hour": 76.16699313759817,
     "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": 1.3684268549986882,
-      "bytes": 11656046,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.130367224002839,
       "bytes": 6468621,
@@ -562,6 +556,12 @@
       "bytes": 7737148,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.883783985998889,
+      "bytes": 16406792,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 761700, "loss": 0.504923403263092, "lr": 5e-05, "grad_norm": 0.23508749085410036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111490271322, "tokens_per_second": 149424.3890953439, "data_wait_s": 9.561001206748188e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761800, "loss": 0.5890706777572632, "lr": 5e-05, "grad_norm": 0.27169946887613655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 152226.03419675006, "data_wait_s": 1.207699824590236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761900, "loss": 0.5406587719917297, "lr": 5e-05, "grad_norm": 0.2462840110372524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111519760426, "tokens_per_second": 152427.16208617773, "data_wait_s": 1.0782998288050294e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 762000, "loss": 0.5618194341659546, "lr": 5e-05, "grad_norm": 0.25727874525994243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111534488578, "tokens_per_second": 152381.51411472692, "data_wait_s": 1.201800478156656e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5137c9658fee19211fc301269988f7321408eb9404a9bc8669bd6d0969c74e91
-size 893871
+oid sha256:725ad9670ddd0ff09b5e1e00d542a8f95471ad55c57f6be6c2b6b40188183920
+size 895989