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": 76980.58360649772,
-      "required_pages_per_second_upper_estimate": 0.005325695393460453
+      "required_token_positions_per_second": 75212.73585646387,
+      "required_pages_per_second_upper_estimate": 0.005203391584141189
     },
     {
       "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": 9622.572950812215,
-      "required_pages_per_second_upper_estimate": 0.004028770352331965
+      "required_token_positions_per_second": 9401.591982057984,
+      "required_pages_per_second_upper_estimate": 0.00393625023532194
     },
     {
       "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": 19245.14590162443,
-      "required_pages_per_second_upper_estimate": 0.004008256362681121
+      "required_token_positions_per_second": 18803.183964115968,
+      "required_pages_per_second_upper_estimate": 0.003916207346418186
     },
     {
       "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": 28867.718852436643,
-      "required_pages_per_second_upper_estimate": 0.0017305756667491143
+      "required_token_positions_per_second": 28204.775946173955,
+      "required_pages_per_second_upper_estimate": 0.0016908332517738727
     },
     {
       "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": 28867.718852436643,
-      "required_pages_per_second_upper_estimate": 0.00224433816510092
+      "required_token_positions_per_second": 28204.775946173955,
+      "required_pages_per_second_upper_estimate": 0.002192797270116671
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28867.718852436643,
-      "required_pages_per_second_upper_estimate": 0.3702841016974724
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28204.775946173955,
+      "required_pages_per_second_upper_estimate": 0.3617805819085691
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4470424027734907,
-    "producer_p95_seconds": 0.30381212399879587,
-    "gpu_update_seconds": 0.9577480001564553,
-    "producer_batches_per_second": 2.23692426891926,
-    "gpu_batches_per_second": 1.0441159885863949,
-    "producer_headroom": 2.142409745058862,
+    "producer_batch_seconds": 0.26527177393751344,
+    "producer_p95_seconds": 0.2841714750029496,
+    "gpu_update_seconds": 0.9802595153658903,
+    "producer_batches_per_second": 3.769718825175711,
+    "gpu_batches_per_second": 1.0201380188865001,
+    "producer_headroom": 3.695302748632416,
     "sustainable": true,
-    "buffer_seconds": 60.338124009856685,
+    "buffer_seconds": 60.7760899526852,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399755.21465372964,
-    "upload_bytes_per_second": 5963845.868077777,
+    "generation_bytes_per_second": 366619.7825668976,
+    "upload_bytes_per_second": 5937081.034411594,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.77816323956411,
+    "estimated_full_shards_per_hour": 78.66807086711117,
     "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": 1.0381844509975053,
-      "bytes": 7741032,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.5200233909999952,
       "bytes": 6160062,
@@ -562,6 +556,12 @@
       "bytes": 13608076,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.8606232550009736,
+      "bytes": 4749519,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -277,3 +277,4 @@
 {"step": 424800, "loss": 0.566508412361145, "lr": 0.0003, "grad_norm": 0.17545006554065132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62179068566, "tokens_per_second": 151526.20415752192, "data_wait_s": 1.5088000509422272e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424900, "loss": 0.5948567390441895, "lr": 0.0003, "grad_norm": 0.1873019857651058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62193678720, "tokens_per_second": 138619.7344759474, "data_wait_s": 1.3924000086262822e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 425000, "loss": 0.5666977763175964, "lr": 0.0003, "grad_norm": 0.14867700949158344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62208440818, "tokens_per_second": 153390.23381304723, "data_wait_s": 1.1512001947266981e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 425100, "loss": 0.5539427399635315, "lr": 0.0003, "grad_norm": 0.16174625622232439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62223279926, "tokens_per_second": 141672.09914388717, "data_wait_s": 1.3191001926315948e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2d6e280f9e1c667585622f919d6d33dccb33c2e1d812e8dd3cfb4cd431f8d4f
-size 595785
+oid sha256:e2baa10bb55baa47edced7432fbd412c2e6af6b2ad23132d63ed394f8c9eeab1
+size 597903