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": 76578.89934217969,
-      "required_pages_per_second_upper_estimate": 0.005297905944019029
+      "required_token_positions_per_second": 76471.21645380359,
+      "required_pages_per_second_upper_estimate": 0.005290456191942445
     },
     {
       "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": 9572.362417772461,
-      "required_pages_per_second_upper_estimate": 0.004007748250663381
+      "required_token_positions_per_second": 9558.902056725448,
+      "required_pages_per_second_upper_estimate": 0.004002112678577298
     },
     {
       "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": 19144.724835544923,
-      "required_pages_per_second_upper_estimate": 0.003987341302898361
+      "required_token_positions_per_second": 19117.804113450897,
+      "required_pages_per_second_upper_estimate": 0.003981734426433359
     },
     {
       "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": 28717.087253317382,
-      "required_pages_per_second_upper_estimate": 0.001721545532383049
+      "required_token_positions_per_second": 28676.706170176345,
+      "required_pages_per_second_upper_estimate": 0.0017191247481070876
     },
     {
       "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": 28717.087253317382,
-      "required_pages_per_second_upper_estimate": 0.002232627220827781
+      "required_token_positions_per_second": 28676.706170176345,
+      "required_pages_per_second_upper_estimate": 0.002229487769231179
     },
     {
       "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": 28717.087253317382,
-      "required_pages_per_second_upper_estimate": 0.3683519612795806
+      "required_token_positions_per_second": 28676.706170176345,
+      "required_pages_per_second_upper_estimate": 0.36783399610287637
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11782661739056266,
-    "producer_p95_seconds": 0.2490375289999065,
-    "gpu_update_seconds": 0.9627717378198277,
-    "producer_batches_per_second": 8.48704666353339,
-    "gpu_batches_per_second": 1.0386677970673244,
-    "producer_headroom": 8.171088665208012,
+    "producer_batch_seconds": 0.1000867255079072,
+    "producer_p95_seconds": 0.23045568999805255,
+    "gpu_update_seconds": 0.9641274641490662,
+    "producer_batches_per_second": 9.991334964005755,
+    "gpu_batches_per_second": 1.0372072544189939,
+    "producer_headroom": 9.632920442310771,
     "sustainable": true,
-    "buffer_seconds": 60.654619482649146,
+    "buffer_seconds": 60.740030241391175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379892.750890103,
-    "upload_bytes_per_second": 5842234.137873605,
+    "generation_bytes_per_second": 379871.26412124926,
+    "upload_bytes_per_second": 5825427.955974279,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.51614089038198,
+    "estimated_full_shards_per_hour": 81.51153032997235,
     "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": 1.444283180999264,
-      "bytes": 6519870,
-      "pages": 22,
-      "shards": 2
-    },
     {
       "seconds": 1.324441224000111,
       "bytes": 10689415,
@@ -562,6 +556,12 @@
       "bytes": 10568739,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9451345529996615,
+      "bytes": 2348924,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -201,3 +201,4 @@
 {"step": 705100, "loss": 0.575244128704071, "lr": 8e-05, "grad_norm": 0.22217285645204765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103206859900, "tokens_per_second": 142009.79687112584, "data_wait_s": 1.3758999557467178e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705200, "loss": 0.522087037563324, "lr": 8e-05, "grad_norm": 0.20907069136539252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103221483343, "tokens_per_second": 152179.45252962134, "data_wait_s": 1.9398001313675195e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705300, "loss": 0.5380771160125732, "lr": 8e-05, "grad_norm": 0.19976465867804705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103236012452, "tokens_per_second": 151303.87789544385, "data_wait_s": 1.5171997802099213e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 705400, "loss": 0.617925226688385, "lr": 8e-05, "grad_norm": 0.23067745386033775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103250575496, "tokens_per_second": 151865.75045943816, "data_wait_s": 1.5477999113500118e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3468e887fc4a87e37370f82d561dcd5756006cd8aea0b8b619c2c37d38bf7895
-size 433251
+oid sha256:73aa3eab943a3faaf0953515376485f2067d85b58350e1702fb34acdbcd8e91a
+size 435369