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": 75509.99152515992,
-      "required_pages_per_second_upper_estimate": 0.0065450146870770064
+      "required_token_positions_per_second": 74831.05790250216,
+      "required_pages_per_second_upper_estimate": 0.006486166441406567
     },
     {
       "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": 9438.74894064499,
-      "required_pages_per_second_upper_estimate": 0.003765258074295911
+      "required_token_positions_per_second": 9353.88223781277,
+      "required_pages_per_second_upper_estimate": 0.003731403477665857
     },
     {
       "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": 18877.49788128998,
-      "required_pages_per_second_upper_estimate": 0.004366976902220879
+      "required_token_positions_per_second": 18707.76447562554,
+      "required_pages_per_second_upper_estimate": 0.004327712066026324
     },
     {
       "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": 28316.24682193497,
-      "required_pages_per_second_upper_estimate": 0.0017243108675903853
+      "required_token_positions_per_second": 28061.64671343831,
+      "required_pages_per_second_upper_estimate": 0.0017088070567664734
     },
     {
       "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": 28316.24682193497,
-      "required_pages_per_second_upper_estimate": 0.002179863064740581
+      "required_token_positions_per_second": 28061.64671343831,
+      "required_pages_per_second_upper_estimate": 0.0021602632436103003
     },
     {
       "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": 28316.24682193497,
-      "required_pages_per_second_upper_estimate": 0.35159301714659064
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28061.64671343831,
+      "required_pages_per_second_upper_estimate": 0.3484317358908118
     }
   ],
   "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": 0.14068510937477186,
-    "producer_p95_seconds": 0.36645837600372033,
-    "gpu_update_seconds": 0.9764005863440449,
-    "producer_batches_per_second": 7.108072804891486,
-    "gpu_batches_per_second": 1.0241698069276246,
-    "producer_headroom": 6.940326454472206,
+    "producer_batch_seconds": 0.1955831520002107,
+    "producer_p95_seconds": 0.3950810390015249,
+    "gpu_update_seconds": 0.985259357098234,
+    "producer_batches_per_second": 5.112914838385071,
+    "gpu_batches_per_second": 1.014961180318226,
+    "producer_headroom": 5.037547186565296,
     "sustainable": true,
-    "buffer_seconds": 60.536836353330784,
+    "buffer_seconds": 60.10082078299227,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272905.2636608797,
-    "upload_bytes_per_second": 5204197.5315556405,
+    "generation_bytes_per_second": 292982.304355705,
+    "upload_bytes_per_second": 5275080.081037062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.559116672227795,
+    "estimated_full_shards_per_hour": 62.86718223574984,
     "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": 0.9286228500022844,
-      "bytes": 849903,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0139027430013812,
       "bytes": 7512493,
@@ -562,6 +556,12 @@
       "bytes": 5614078,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.196396844003175,
+      "bytes": 7831133,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -366,3 +366,4 @@
 {"step": 273700, "loss": 0.5720657892525196, "lr": 0.0003, "grad_norm": 0.15198303023170256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40060943559, "tokens_per_second": 150887.69928129172, "data_wait_s": 9.119998139794916e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 273800, "loss": 0.6430566664785147, "lr": 0.0003, "grad_norm": 0.16033061241497748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40075615549, "tokens_per_second": 148455.10996529672, "data_wait_s": 1.4830002328380942e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273900, "loss": 0.5698623228818178, "lr": 0.0003, "grad_norm": 0.13981164257315393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40090351142, "tokens_per_second": 150803.33100614074, "data_wait_s": 1.2078002328053117e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 274000, "loss": 0.5780625939369202, "lr": 0.0003, "grad_norm": 0.14163760091911487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40105112994, "tokens_per_second": 149696.021686254, "data_wait_s": 1.6744001186452806e-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:dbdf5b41f8e13ebc2f07339dec2ef07cadc786dffaf5bd7532fb5bb44a39f8b7
-size 785070
+oid sha256:de6f9456f4ac8c51f3577aa5363255dc22981f60c080e7891bada09d70f2fed8
+size 787188