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": 75688.48203985496,
-      "required_pages_per_second_upper_estimate": 0.0065604857925107
+      "required_token_positions_per_second": 75419.84210860054,
+      "required_pages_per_second_upper_estimate": 0.006537200764131247
     },
     {
       "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": 9461.06025498187,
-      "required_pages_per_second_upper_estimate": 0.003774158391168769
+      "required_token_positions_per_second": 9427.480263575068,
+      "required_pages_per_second_upper_estimate": 0.003760762830530983
     },
     {
       "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": 18922.12050996374,
-      "required_pages_per_second_upper_estimate": 0.004377299562033111
+      "required_token_positions_per_second": 18854.960527150135,
+      "required_pages_per_second_upper_estimate": 0.0043617632819845115
     },
     {
       "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": 28383.180764945613,
-      "required_pages_per_second_upper_estimate": 0.0017283867935444749
+      "required_token_positions_per_second": 28282.440790725203,
+      "required_pages_per_second_upper_estimate": 0.0017222522576562488
     },
     {
       "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": 28383.180764945613,
-      "required_pages_per_second_upper_estimate": 0.0021850158249585535
+      "required_token_positions_per_second": 28282.440790725203,
+      "required_pages_per_second_upper_estimate": 0.0021772605828769692
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28383.180764945613,
-      "required_pages_per_second_upper_estimate": 0.35242411270528595
+      "required_token_positions_per_second": 28282.440790725203,
+      "required_pages_per_second_upper_estimate": 0.3511732593804736
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08876994418767481,
-    "producer_p95_seconds": 0.25612042699867743,
-    "gpu_update_seconds": 0.9740980135019401,
-    "producer_batches_per_second": 11.265074109834172,
-    "gpu_batches_per_second": 1.0265907394728593,
-    "producer_headroom": 10.973286312341605,
+    "producer_batch_seconds": 0.12100791143726042,
+    "producer_p95_seconds": 0.27829360000032466,
+    "gpu_update_seconds": 0.9775676789913671,
+    "producer_batches_per_second": 8.263922483435929,
+    "gpu_batches_per_second": 1.0229470772108362,
+    "producer_headroom": 8.078543521497036,
     "sustainable": true,
-    "buffer_seconds": 60.39407683712029,
+    "buffer_seconds": 60.60919609746476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 249106.8227141804,
-    "upload_bytes_per_second": 5319634.948624714,
+    "generation_bytes_per_second": 260984.3657879055,
+    "upload_bytes_per_second": 5229629.781649752,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.45252524441775,
+    "estimated_full_shards_per_hour": 56.00116949298738,
     "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": 1.0392706679995172,
-      "bytes": 8238622,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0565556260007725,
       "bytes": 5902213,
@@ -562,6 +556,12 @@
       "bytes": 10407908,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.2640827479990548,
+      "bytes": 2266301,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 256700, "loss": 0.5793802235275507, "lr": 0.0003, "grad_norm": 0.1362063822111426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37573826960, "tokens_per_second": 149937.26378596557, "data_wait_s": 1.3355998817132786e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256800, "loss": 0.5809282548725605, "lr": 0.0003, "grad_norm": 0.16060970723430443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37588422835, "tokens_per_second": 125301.86413422806, "data_wait_s": 1.4684002962894738e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256900, "loss": 0.5757004469633102, "lr": 0.0003, "grad_norm": 0.15516702809483268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37602925941, "tokens_per_second": 147485.50615365646, "data_wait_s": 1.4310997357824817e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 257000, "loss": 0.6060768719762564, "lr": 0.0003, "grad_norm": 0.17568757799964974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37617394237, "tokens_per_second": 146889.6447210031, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7071d2879ebb41eac534d3a305ce36cfa3a9742d1f744949e7cc625673d2698e
-size 422661
+oid sha256:f0a7a395c29d9df5949e70b77a0cd0fbb5a0b46ca4f98537d0f0b8e1d7a50f12
+size 424779