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": 75576.09724670161,
-      "required_pages_per_second_upper_estimate": 0.00522852976823242
+      "required_token_positions_per_second": 77275.6032103166,
+      "required_pages_per_second_upper_estimate": 0.005346105534192435
     },
     {
       "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": 9447.012155837701,
-      "required_pages_per_second_upper_estimate": 0.003955266713602424
+      "required_token_positions_per_second": 9659.450401289576,
+      "required_pages_per_second_upper_estimate": 0.004044210170758101
     },
     {
       "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": 18894.024311675403,
-      "required_pages_per_second_upper_estimate": 0.0039351269951937535
+      "required_token_positions_per_second": 19318.90080257915,
+      "required_pages_per_second_upper_estimate": 0.004023617563502452
     },
     {
       "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": 28341.036467513102,
-      "required_pages_per_second_upper_estimate": 0.0016990018619703854
+      "required_token_positions_per_second": 28978.35120386873,
+      "required_pages_per_second_upper_estimate": 0.0017372079072916484
     },
     {
       "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": 28341.036467513102,
-      "required_pages_per_second_upper_estimate": 0.002203390926304097
+      "required_token_positions_per_second": 28978.35120386873,
+      "required_pages_per_second_upper_estimate": 0.0022529393438044786
     },
     {
       "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": 28341.036467513102,
-      "required_pages_per_second_upper_estimate": 0.3635283855711587
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28978.35120386873,
+      "required_pages_per_second_upper_estimate": 0.3717031747138791
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3686582535472098,
-    "producer_p95_seconds": 0.3421674630008056,
-    "gpu_update_seconds": 0.9755465376748829,
-    "producer_batches_per_second": 2.712539297243596,
-    "gpu_batches_per_second": 1.0250664231594728,
-    "producer_headroom": 2.64620831973305,
+    "producer_batch_seconds": 0.10766386294551467,
+    "producer_p95_seconds": 0.32867089600040345,
+    "gpu_update_seconds": 0.9540915494291089,
+    "producer_batches_per_second": 9.28816756747869,
+    "gpu_batches_per_second": 1.0481174480565947,
+    "producer_headroom": 8.86176218581294,
     "sustainable": true,
-    "buffer_seconds": 60.483885335842736,
+    "buffer_seconds": 60.10776761403386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 533874.956817281,
-    "upload_bytes_per_second": 5968492.4491932755,
+    "generation_bytes_per_second": 535253.9099393168,
+    "upload_bytes_per_second": 5961717.847362944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.55713776005575,
+    "estimated_full_shards_per_hour": 114.85302899965885,
     "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": 1.1199947279965272,
-      "bytes": 7293697,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0610679239980527,
       "bytes": 5725336,
@@ -562,6 +556,12 @@
       "bytes": 18139508,
       "pages": 43,
       "shards": 5
+    },
+    {
+      "seconds": 0.9787014109970187,
+      "bytes": 5956252,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 516300, "loss": 0.6054825186729431, "lr": 0.0003, "grad_norm": 0.15628910358101797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75568475628, "tokens_per_second": 151586.38665990692, "data_wait_s": 1.6095997125376016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516400, "loss": 0.5965888500213623, "lr": 0.0003, "grad_norm": 0.1514518811000016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75583191886, "tokens_per_second": 153272.14447573066, "data_wait_s": 2.549700002418831e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516500, "loss": 0.5942535400390625, "lr": 0.0003, "grad_norm": 0.16532523981230376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75597973591, "tokens_per_second": 153758.60466063625, "data_wait_s": 2.15519976336509e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 516600, "loss": 0.5852409601211548, "lr": 0.0003, "grad_norm": 0.16251029845161472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75612706432, "tokens_per_second": 153456.53370417174, "data_wait_s": 1.7587997717782855e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b9a956d9c3078bde62ba918da7b0c0997d06f68a569d5b0590bdc15be9409e3
-size 672033
+oid sha256:3e1329b59007b3af0dc068254a6efe8d4d71b49792891005ce382182579b803e
+size 674151