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": 75492.62586324968,
-      "required_pages_per_second_upper_estimate": 0.0065435094755154945
+      "required_token_positions_per_second": 73864.39236615006,
+      "required_pages_per_second_upper_estimate": 0.00640237832270697
     },
     {
       "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": 9436.57823290621,
-      "required_pages_per_second_upper_estimate": 0.0037643921465239385
+      "required_token_positions_per_second": 9233.049045768757,
+      "required_pages_per_second_upper_estimate": 0.003683201310742284
     },
     {
       "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": 18873.15646581242,
-      "required_pages_per_second_upper_estimate": 0.004365972592156448
+      "required_token_positions_per_second": 18466.098091537515,
+      "required_pages_per_second_upper_estimate": 0.004271806801250656
     },
     {
       "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": 28309.734698718632,
-      "required_pages_per_second_upper_estimate": 0.0017239143134529806
+      "required_token_positions_per_second": 27699.147137306274,
+      "required_pages_per_second_upper_estimate": 0.0016867327344683228
     },
     {
       "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": 28309.734698718632,
-      "required_pages_per_second_upper_estimate": 0.0021793617434686767
+      "required_token_positions_per_second": 27699.147137306274,
+      "required_pages_per_second_upper_estimate": 0.002132357022776605
     },
     {
       "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": 28309.734698718632,
-      "required_pages_per_second_upper_estimate": 0.3515121583709181
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27699.147137306274,
+      "required_pages_per_second_upper_estimate": 0.3439307043632898
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1185122089922288,
-    "producer_p95_seconds": 0.39575560000230325,
-    "gpu_update_seconds": 0.9766251889761234,
-    "producer_batches_per_second": 8.437949207963653,
-    "gpu_batches_per_second": 1.0239342700636078,
-    "producer_headroom": 8.240713739798434,
+    "producer_batch_seconds": 0.10240371547678251,
+    "producer_p95_seconds": 0.3902716260054149,
+    "gpu_update_seconds": 0.9981534760961146,
+    "producer_batches_per_second": 9.765270677378156,
+    "gpu_batches_per_second": 1.0018499398620615,
+    "producer_headroom": 9.747238871644468,
     "sustainable": true,
-    "buffer_seconds": 60.55076171651965,
+    "buffer_seconds": 60.88736204186299,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382873.19731244753,
-    "upload_bytes_per_second": 3782178.175520555,
+    "generation_bytes_per_second": 419581.2877037805,
+    "upload_bytes_per_second": 3764420.293481104,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.15567531137533,
+    "estimated_full_shards_per_hour": 90.03237698874533,
     "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": 0.9533051110011002,
-      "bytes": 7632407,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.273418650001986,
       "bytes": 11058609,
@@ -562,6 +556,12 @@
       "bytes": 7400872,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1047442850031075,
+      "bytes": 6203278,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 38500, "loss": 0.6896477043628693, "lr": 0.0003, "grad_norm": 0.1378860124558519, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5633848010, "tokens_per_second": 148474.49969914253, "data_wait_s": 1.6270001651719213e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38600, "loss": 0.6759868860244751, "lr": 0.0003, "grad_norm": 0.12844563084328195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5648475620, "tokens_per_second": 148436.1762926683, "data_wait_s": 1.2041004083584994e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38700, "loss": 0.6552939936518669, "lr": 0.0003, "grad_norm": 0.12835237871211294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5663203830, "tokens_per_second": 149862.03059944644, "data_wait_s": 9.670999133959413e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 38800, "loss": 0.6750294882804155, "lr": 0.0003, "grad_norm": 0.15139465182627893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5677886427, "tokens_per_second": 148717.4995696666, "data_wait_s": 1.4072000340092927e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c8438a7c2e575ac1c7ea4d79da87fada9c7cb75f30da19565b75ebaed7e9d30
-size 819538
+oid sha256:d4412afd2e3b715d56d8e236148b8e00b432f1f5913aac60d145a954f7471234
+size 821656