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": 75490.84711997952,
-      "required_pages_per_second_upper_estimate": 0.006543355298557021
+      "required_token_positions_per_second": 75503.2388063348,
+      "required_pages_per_second_upper_estimate": 0.0065444293785238555
     },
     {
       "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": 9436.35588999744,
-      "required_pages_per_second_upper_estimate": 0.0037643034506133076
+      "required_token_positions_per_second": 9437.90485079185,
+      "required_pages_per_second_upper_estimate": 0.0037649213542332257
     },
     {
       "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": 18872.71177999488,
-      "required_pages_per_second_upper_estimate": 0.004365869721918763
+      "required_token_positions_per_second": 18875.8097015837,
+      "required_pages_per_second_upper_estimate": 0.00436658637155678
     },
     {
       "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": 28309.06766999232,
-      "required_pages_per_second_upper_estimate": 0.0017238736949031774
+      "required_token_positions_per_second": 28313.71455237555,
+      "required_pages_per_second_upper_estimate": 0.0017241566656600084
     },
     {
       "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": 28309.06766999232,
-      "required_pages_per_second_upper_estimate": 0.0021793103937508716
+      "required_token_positions_per_second": 28313.71455237555,
+      "required_pages_per_second_upper_estimate": 0.0021796681236201263
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28309.06766999232,
-      "required_pages_per_second_upper_estimate": 0.3515038761065389
+      "required_token_positions_per_second": 28313.71455237555,
+      "required_pages_per_second_upper_estimate": 0.35156157483362366
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05445986976554451,
-    "producer_p95_seconds": 0.18568832499931887,
-    "gpu_update_seconds": 0.9766482005801607,
-    "producer_batches_per_second": 18.362144535143134,
-    "gpu_batches_per_second": 1.0239101443139582,
-    "producer_headroom": 17.933355419040375,
+    "producer_batch_seconds": 0.07835912122659039,
+    "producer_p95_seconds": 0.17468036499940354,
+    "gpu_update_seconds": 0.9764879118511952,
+    "producer_batches_per_second": 12.761756185451706,
+    "gpu_batches_per_second": 1.0240782173168241,
+    "producer_headroom": 12.461700649085811,
     "sustainable": true,
-    "buffer_seconds": 60.55218843596996,
+    "buffer_seconds": 60.542250534774105,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341184.63619351597,
+    "generation_bytes_per_second": 375331.1197161932,
     "upload_bytes_per_second": 5441680.69509299,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.21028055528745,
+    "estimated_full_shards_per_hour": 80.5373210297999,
     "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": [

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

@@ -54,3 +54,4 @@
 {"step": 120400, "loss": 0.5748705342411995, "lr": 0.0003, "grad_norm": 0.15546136931297286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17619138984, "tokens_per_second": 147675.25254648548, "data_wait_s": 1.7296999430982396e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120500, "loss": 0.6057937908917665, "lr": 0.0003, "grad_norm": 0.1277409518869648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17633614354, "tokens_per_second": 148114.15318738951, "data_wait_s": 1.4208999346010387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120600, "loss": 0.5935172829777002, "lr": 0.0003, "grad_norm": 0.13671291106910105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17648295800, "tokens_per_second": 150081.69821159675, "data_wait_s": 9.203999979945365e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 120700, "loss": 0.6122833862900734, "lr": 0.0003, "grad_norm": 0.14585721939975288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17663007237, "tokens_per_second": 150323.7249672456, "data_wait_s": 1.301300017075846e-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 sha256:2aa9157479e043c86909ee49517511566d4784eb9f8ddb2eb10e7467b5f3b0e3
-size 119556
+oid sha256:ff4d96b386830b357f77f19268ed0d8ef74a4feef931c860c2fa84844bd33b55
+size 121674