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": 75378.47436775688,
-      "required_pages_per_second_upper_estimate": 0.006533615113200499
+      "required_token_positions_per_second": 75099.7865566571,
+      "required_pages_per_second_upper_estimate": 0.0065094591600621705
     },
     {
       "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": 9422.30929596961,
-      "required_pages_per_second_upper_estimate": 0.0037587000542403104
+      "required_token_positions_per_second": 9387.473319582137,
+      "required_pages_per_second_upper_estimate": 0.003744803462415086
     },
     {
       "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": 18844.61859193922,
-      "required_pages_per_second_upper_estimate": 0.004359370857285309
+      "required_token_positions_per_second": 18774.946639164275,
+      "required_pages_per_second_upper_estimate": 0.004343253477195315
     },
     {
       "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": 28266.92788790883,
-      "required_pages_per_second_upper_estimate": 0.0017213076032646427
+      "required_token_positions_per_second": 28162.419958746414,
+      "required_pages_per_second_upper_estimate": 0.0017149436186893801
     },
     {
       "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": 28266.92788790883,
-      "required_pages_per_second_upper_estimate": 0.0021760663566757015
+      "required_token_positions_per_second": 28162.419958746414,
+      "required_pages_per_second_upper_estimate": 0.002168021046992315
     },
     {
       "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": 28266.92788790883,
-      "required_pages_per_second_upper_estimate": 0.3509806410458402
+      "required_token_positions_per_second": 28162.419958746414,
+      "required_pages_per_second_upper_estimate": 0.3496830023311821
     }
   ],
   "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": 0.09097722060920432,
-    "producer_p95_seconds": 0.22313243000098737,
-    "gpu_update_seconds": 0.9781041685759712,
-    "producer_batches_per_second": 10.991762479703938,
-    "gpu_batches_per_second": 1.022385991316147,
-    "producer_headroom": 10.751088701395377,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.64245845171021,
+    "buffer_seconds": 60.867496561410654,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303256.2198315552,
-    "upload_bytes_per_second": 5275077.508998218,
+    "generation_bytes_per_second": 273311.4166933994,
+    "upload_bytes_per_second": 5353793.684818373,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.07172533235543,
+    "estimated_full_shards_per_hour": 58.64626765824782,
     "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": 1.183389565001562,
-      "bytes": 6316614,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.805858101000922,
       "bytes": 7507792,
@@ -562,6 +556,12 @@
       "bytes": 2406820,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0375558390005608,
+      "bytes": 11763695,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"step": 257200, "loss": 0.613028310239315, "lr": 0.0003, "grad_norm": 0.14739748755808474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37646738818, "tokens_per_second": 149730.20846703582, "data_wait_s": 1.3818000297760591e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:939d024cb3571e47152f0f457a20545b15153a3b3c16386e4abcb05a4a13835e
-size 433251
+oid sha256:66d79c908017b8b11b081fab03ee1cb735e5dd5cc9c6f651ed06f24b62369050
+size 435369