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": 76302.00751205217,
-      "required_pages_per_second_upper_estimate": 0.005278749924733244
+      "required_token_positions_per_second": 76361.73089873877,
+      "required_pages_per_second_upper_estimate": 0.005282881727201573
     },
     {
       "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": 9537.75093900652,
-      "required_pages_per_second_upper_estimate": 0.0039932571472739474
+      "required_token_positions_per_second": 9545.216362342346,
+      "required_pages_per_second_upper_estimate": 0.00399638276412889
     },
     {
       "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": 19075.50187801304,
-      "required_pages_per_second_upper_estimate": 0.0039729239863766225
+      "required_token_positions_per_second": 19090.43272468469,
+      "required_pages_per_second_upper_estimate": 0.003976033687985422
     },
     {
       "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": 28613.25281701956,
-      "required_pages_per_second_upper_estimate": 0.0017153208164729986
+      "required_token_positions_per_second": 28635.64908702704,
+      "required_pages_per_second_upper_estimate": 0.0017166634386628229
     },
     {
       "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": 28613.25281701956,
-      "required_pages_per_second_upper_estimate": 0.002224554550125043
+      "required_token_positions_per_second": 28635.64908702704,
+      "required_pages_per_second_upper_estimate": 0.0022262957616073437
     },
     {
       "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": 28613.25281701956,
-      "required_pages_per_second_upper_estimate": 0.36702008461948354
+      "required_token_positions_per_second": 28635.64908702704,
+      "required_pages_per_second_upper_estimate": 0.36730735992389835
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1157894923047138,
-    "producer_p95_seconds": 0.21753807900131505,
-    "gpu_update_seconds": 0.9662655335556461,
-    "producer_batches_per_second": 8.636362247520537,
-    "gpu_batches_per_second": 1.0349122112637283,
-    "producer_headroom": 8.345019175080271,
+    "producer_batch_seconds": 0.0558197019609139,
+    "producer_p95_seconds": 0.1932843659997161,
+    "gpu_update_seconds": 0.9655098061850996,
+    "producer_batches_per_second": 17.914821557095028,
+    "gpu_batches_per_second": 1.035722261538883,
+    "producer_headroom": 17.296935889431467,
     "sustainable": true,
-    "buffer_seconds": 60.874728614005704,
+    "buffer_seconds": 60.82711778966128,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355120.98886369186,
-    "upload_bytes_per_second": 4049920.551415638,
+    "generation_bytes_per_second": 387383.5001390832,
+    "upload_bytes_per_second": 4062565.0332068633,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.2006974166209,
+    "estimated_full_shards_per_hour": 83.12348130349514,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 10891963,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.62565908400029,
+      "bytes": 7724112,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 810600, "loss": 0.538456380367279, "lr": 3e-05, "grad_norm": 0.276272048776533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118648949202, "tokens_per_second": 139873.5227415592, "data_wait_s": 1.6869000319275074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 810700, "loss": 0.5303323864936829, "lr": 3e-05, "grad_norm": 0.2866013259913802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118663650275, "tokens_per_second": 151936.2575958504, "data_wait_s": 1.2083999536116607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 810800, "loss": 0.5498128533363342, "lr": 3e-05, "grad_norm": 0.27593817248432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118678180898, "tokens_per_second": 150128.4985404471, "data_wait_s": 1.633300053072162e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 810900, "loss": 0.5189882516860962, "lr": 3e-05, "grad_norm": 0.2833655514270679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118692713724, "tokens_per_second": 150418.51877316384, "data_wait_s": 1.1419000657042488e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9146a4d058f65c04b0056f17c3adc1e462ce6f43c61d316db9f390a1267f279b
-size 230475
+oid sha256:8495aa014e5c26b83aee6b596e99e36c4f442a95bfdcae1600c651fd2615162d
+size 232593