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": 75099.7865566571,
-      "required_pages_per_second_upper_estimate": 0.0065094591600621705
+      "required_token_positions_per_second": 74164.52710767383,
+      "required_pages_per_second_upper_estimate": 0.006428393241417703
     },
     {
       "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": 9387.473319582137,
-      "required_pages_per_second_upper_estimate": 0.003744803462415086
+      "required_token_positions_per_second": 9270.565888459229,
+      "required_pages_per_second_upper_estimate": 0.0036981673402182976
     },
     {
       "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": 18774.946639164275,
-      "required_pages_per_second_upper_estimate": 0.004343253477195315
+      "required_token_positions_per_second": 18541.131776918457,
+      "required_pages_per_second_upper_estimate": 0.004289164523815776
     },
     {
       "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": 28162.419958746414,
-      "required_pages_per_second_upper_estimate": 0.0017149436186893801
+      "required_token_positions_per_second": 27811.697665377687,
+      "required_pages_per_second_upper_estimate": 0.0016935864711208877
     },
     {
       "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": 28162.419958746414,
-      "required_pages_per_second_upper_estimate": 0.002168021046992315
+      "required_token_positions_per_second": 27811.697665377687,
+      "required_pages_per_second_upper_estimate": 0.0021410214739873445
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28162.419958746414,
-      "required_pages_per_second_upper_estimate": 0.3496830023311821
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27811.697665377687,
+      "required_pages_per_second_upper_estimate": 0.3453282052395506
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.40434101376567355,
-    "producer_p95_seconds": 0.28577520299950265,
-    "gpu_update_seconds": 0.9817338155066234,
-    "producer_batches_per_second": 2.4731599465680887,
-    "gpu_batches_per_second": 1.0186060459616035,
-    "producer_headroom": 2.427984750702447,
+    "producer_batch_seconds": 0.12314553904707282,
+    "producer_p95_seconds": 0.2824703660007799,
+    "gpu_update_seconds": 0.9941140714476603,
+    "producer_batches_per_second": 8.120472797782359,
+    "gpu_batches_per_second": 1.0059207778276074,
+    "producer_headroom": 8.072676275083392,
     "sustainable": true,
-    "buffer_seconds": 60.867496561410654,
+    "buffer_seconds": 60.64095835830728,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273311.4166933994,
-    "upload_bytes_per_second": 5353793.684818373,
+    "generation_bytes_per_second": 285694.1471566007,
+    "upload_bytes_per_second": 5309280.636101666,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.64626765824782,
+    "estimated_full_shards_per_hour": 61.3033133604385,
     "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": 1.805858101000922,
-      "bytes": 7507792,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 2.5084295430006023,
       "bytes": 11998594,
@@ -562,6 +556,12 @@
       "bytes": 11763695,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.2267565980000654,
+      "bytes": 917900,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -202,3 +202,4 @@
 {"step": 257300, "loss": 0.6018412634730339, "lr": 0.0003, "grad_norm": 0.13902728863561256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37661449005, "tokens_per_second": 149241.48874367544, "data_wait_s": 1.1223997717024758e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 257400, "loss": 0.5995901804417372, "lr": 0.0003, "grad_norm": 0.139070939425421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37675984408, "tokens_per_second": 147966.88766195363, "data_wait_s": 9.192001016344875e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 257500, "loss": 0.5935665965080261, "lr": 0.0003, "grad_norm": 0.1467446045679039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37690660025, "tokens_per_second": 149214.93510628436, "data_wait_s": 1.3618999219033867e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 257600, "loss": 0.5662095006555319, "lr": 0.0003, "grad_norm": 0.15324470293276973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37705172210, "tokens_per_second": 147290.96547301702, "data_wait_s": 1.5392000932479277e-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 sha256:66d79c908017b8b11b081fab03ee1cb735e5dd5cc9c6f651ed06f24b62369050
-size 435369
+oid sha256:04a7c089c022b511a9ed658d6401b9f07366b65bf2bcdd2727dc69ebde6e2046
+size 437487