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": 77359.81711174686,
-      "required_pages_per_second_upper_estimate": 0.005351931647296553
+      "required_token_positions_per_second": 77492.71061061385,
+      "required_pages_per_second_upper_estimate": 0.005361125527903572
     },
     {
       "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": 9669.977138968357,
-      "required_pages_per_second_upper_estimate": 0.004048617496000927
+      "required_token_positions_per_second": 9686.588826326732,
+      "required_pages_per_second_upper_estimate": 0.0040555724626064
     },
     {
       "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": 19339.954277936715,
-      "required_pages_per_second_upper_estimate": 0.004028002447201951
+      "required_token_positions_per_second": 19373.177652653463,
+      "required_pages_per_second_upper_estimate": 0.004034921999996139
     },
     {
       "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": 29009.931416905074,
-      "required_pages_per_second_upper_estimate": 0.001739101092842984
+      "required_token_positions_per_second": 29059.766478980193,
+      "required_pages_per_second_upper_estimate": 0.0017420886287206536
     },
     {
       "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": 29009.931416905074,
-      "required_pages_per_second_upper_estimate": 0.0022553945664613773
+      "required_token_positions_per_second": 29059.766478980193,
+      "required_pages_per_second_upper_estimate": 0.0022592690233364456
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29009.931416905074,
-      "required_pages_per_second_upper_estimate": 0.3721082517785184
+      "required_token_positions_per_second": 29059.766478980193,
+      "required_pages_per_second_upper_estimate": 0.3727474824460973
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10705924866406491,
-    "producer_p95_seconds": 0.18382348799968895,
-    "gpu_update_seconds": 0.9530529253126249,
-    "producer_batches_per_second": 9.340622248693737,
-    "gpu_batches_per_second": 1.0492596721970875,
-    "producer_headroom": 8.902107358357755,
+    "producer_batch_seconds": 0.11088183010155461,
+    "producer_p95_seconds": 0.18175020000035147,
+    "gpu_update_seconds": 0.9514185194846158,
+    "producer_batches_per_second": 9.018610164389589,
+    "gpu_batches_per_second": 1.0510621556344109,
+    "producer_headroom": 8.58047273041245,
     "sustainable": true,
-    "buffer_seconds": 60.042334294695365,
+    "buffer_seconds": 60.89078524701541,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327854.99902186304,
-    "upload_bytes_per_second": 5161892.294772752,
+    "generation_bytes_per_second": 314407.2346590987,
+    "upload_bytes_per_second": 5197019.741949496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.35005071632307,
+    "estimated_full_shards_per_hour": 67.46447353200647,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 5176693,
       "pages": 44,
       "shards": 2
+    },
+    {
+      "seconds": 1.123353902999952,
+      "bytes": 7475982,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 890700, "loss": 0.5632334351539612, "lr": 3e-05, "grad_norm": 0.31817407662024405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130373755907, "tokens_per_second": 142281.99469000564, "data_wait_s": 9.66199968388537e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 890800, "loss": 0.5203015804290771, "lr": 3e-05, "grad_norm": 0.3147710947317427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130388333487, "tokens_per_second": 151918.67317398838, "data_wait_s": 1.0480999662831891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890900, "loss": 0.5388245582580566, "lr": 3e-05, "grad_norm": 0.3211956049093675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130402871994, "tokens_per_second": 151829.41128575586, "data_wait_s": 1.174099998024758e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 891000, "loss": 0.5289643406867981, "lr": 3e-05, "grad_norm": 0.3304750209118958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130417428723, "tokens_per_second": 137786.4988271708, "data_wait_s": 1.584900019224733e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bc555bd0fc89f35f089e8ba48cf9bee4d106d5e02d996b47c8ea27b690e0c87
-size 125910
+oid sha256:6aa67b9091c5f78e5e18b028d49162776103c415ea02345d22d61662679c2551
+size 128028