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": 77150.56825100051,
-      "required_pages_per_second_upper_estimate": 0.005337455325585871
+      "required_token_positions_per_second": 77640.74032217583,
+      "required_pages_per_second_upper_estimate": 0.005371366566825682
     },
     {
       "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": 9643.821031375064,
-      "required_pages_per_second_upper_estimate": 0.004037666479953252
+      "required_token_positions_per_second": 9705.092540271979,
+      "required_pages_per_second_upper_estimate": 0.004063319581233788
     },
     {
       "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": 19287.64206275013,
-      "required_pages_per_second_upper_estimate": 0.004017107192344472
+      "required_token_positions_per_second": 19410.185080543957,
+      "required_pages_per_second_upper_estimate": 0.004042629671274226
     },
     {
       "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": 28931.46309412519,
-      "required_pages_per_second_upper_estimate": 0.0017343970367065214
+      "required_token_positions_per_second": 29115.277620815934,
+      "required_pages_per_second_upper_estimate": 0.0017454164369131004
     },
     {
       "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": 28931.46309412519,
-      "required_pages_per_second_upper_estimate": 0.002249293999510917
+      "required_token_positions_per_second": 29115.277620815934,
+      "required_pages_per_second_upper_estimate": 0.002263584770446466
     },
     {
       "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": 28931.46309412519,
-      "required_pages_per_second_upper_estimate": 0.3711017443866188
+      "required_token_positions_per_second": 29115.277620815934,
+      "required_pages_per_second_upper_estimate": 0.37345951977034586
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06641551042210381,
-    "producer_p95_seconds": 0.21092751500327722,
-    "gpu_update_seconds": 0.9556378089158647,
-    "producer_batches_per_second": 15.056723853276132,
-    "gpu_batches_per_second": 1.046421552883579,
-    "producer_headroom": 14.38877459259604,
+    "producer_batch_seconds": 0.11699731936741387,
+    "producer_p95_seconds": 0.23951043800116167,
+    "gpu_update_seconds": 0.9496045464540958,
+    "producer_batches_per_second": 8.547204375338195,
+    "gpu_batches_per_second": 1.0530699370954837,
+    "producer_headroom": 8.11646413429349,
     "sustainable": true,
-    "buffer_seconds": 60.20518196169947,
+    "buffer_seconds": 60.77469097306213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279924.7576734005,
-    "upload_bytes_per_second": 5120329.741592116,
+    "generation_bytes_per_second": 280156.71588639513,
+    "upload_bytes_per_second": 5013754.580396921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.06533668185722,
+    "estimated_full_shards_per_hour": 60.115109514655025,
     "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": 0.9986490080009389,
-      "bytes": 2653371,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1648389490001136,
       "bytes": 5951393,
@@ -562,6 +556,12 @@
       "bytes": 11141641,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 3.434384232998127,
+      "bytes": 5971898,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -170,3 +170,4 @@
 {"step": 902000, "loss": 0.5099591016769409, "lr": 3e-05, "grad_norm": 0.31409916288570733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132026993992, "tokens_per_second": 134465.43752851879, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 902100, "loss": 0.5030158162117004, "lr": 3e-05, "grad_norm": 0.3099639529913761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132041691282, "tokens_per_second": 152593.38545743533, "data_wait_s": 1.7455000488553196e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 902200, "loss": 0.5325385332107544, "lr": 3e-05, "grad_norm": 0.3249382554368432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132056366016, "tokens_per_second": 153416.69024096013, "data_wait_s": 1.0091000149259344e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 902300, "loss": 0.5309786200523376, "lr": 3e-05, "grad_norm": 0.3312332507631469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132070973656, "tokens_per_second": 150914.83163535254, "data_wait_s": 1.4658999134553596e-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:d7fa8c295895b04cc2a4cb439a85c4ed7c836c2076f359f811a39f6a75a65b92
-size 366810
+oid sha256:12a05454bc78a3ea32e47820095a6734cdfbf9faab77b67fff9405978f80ccfa
+size 368928