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": 75120.45760388167,
-      "required_pages_per_second_upper_estimate": 0.006511250874045306
+      "required_token_positions_per_second": 74810.88023184256,
+      "required_pages_per_second_upper_estimate": 0.0064844174920536315
     },
     {
       "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": 9390.05720048521,
-      "required_pages_per_second_upper_estimate": 0.0037458342111397836
+      "required_token_positions_per_second": 9351.36002898032,
+      "required_pages_per_second_upper_estimate": 0.00373039733085221
     },
     {
       "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": 18780.11440097042,
-      "required_pages_per_second_upper_estimate": 0.004344448947939665
+      "required_token_positions_per_second": 18702.72005796064,
+      "required_pages_per_second_upper_estimate": 0.004326545128778269
     },
     {
       "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": 28170.17160145563,
-      "required_pages_per_second_upper_estimate": 0.0017154156530606451
+      "required_token_positions_per_second": 28054.080086940958,
+      "required_pages_per_second_upper_estimate": 0.0017083462889118058
     },
     {
       "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": 28170.17160145563,
-      "required_pages_per_second_upper_estimate": 0.0021686177899059907
+      "required_token_positions_per_second": 28054.080086940958,
+      "required_pages_per_second_upper_estimate": 0.002159680743756771
     },
     {
       "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": 28170.17160145563,
-      "required_pages_per_second_upper_estimate": 0.3497792517905513
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28054.080086940958,
+      "required_pages_per_second_upper_estimate": 0.34833778371358454
     }
   ],
   "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": 0.20651756988294778,
-    "producer_p95_seconds": 0.31602960099917254,
-    "gpu_update_seconds": 0.9814636698404546,
-    "producer_batches_per_second": 4.842203017238633,
-    "gpu_batches_per_second": 1.0188864149832042,
-    "producer_headroom": 4.752446343411551,
+    "producer_batch_seconds": 0.07825088955476645,
+    "producer_p95_seconds": 0.3287395330007712,
+    "gpu_update_seconds": 0.9855250970382028,
+    "producer_batches_per_second": 12.779407437919504,
+    "gpu_batches_per_second": 1.014687503144566,
+    "producer_headroom": 12.594426755346351,
     "sustainable": true,
-    "buffer_seconds": 60.85074753010819,
+    "buffer_seconds": 60.11703091933037,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 628791.9512074635,
-    "upload_bytes_per_second": 2651314.957309086,
+    "generation_bytes_per_second": 633201.8261330775,
+    "upload_bytes_per_second": 2674755.2860557325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.4620576008221,
+    "estimated_full_shards_per_hour": 67.93518585202334,
     "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": 1.3497758060002525,
-      "bytes": 7313270,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.170668664999539,
       "bytes": 6087923,
@@ -562,6 +556,12 @@
       "bytes": 2626145,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0629035870006192,
+      "bytes": 10213495,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 299800, "loss": 0.5831612106412649, "lr": 0.0003, "grad_norm": 0.1823047531417891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43883277447, "tokens_per_second": 150319.82842850953, "data_wait_s": 1.3925000530434772e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 299900, "loss": 0.6247671563178301, "lr": 0.0003, "grad_norm": 0.18420449801814362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43898006866, "tokens_per_second": 150604.85477110342, "data_wait_s": 9.949999366654083e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 300000, "loss": 0.6110201980918646, "lr": 0.0003, "grad_norm": 0.15293899767396202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43912525103, "tokens_per_second": 148928.3343444987, "data_wait_s": 1.3712997315451503e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 300100, "loss": 0.5722170490771532, "lr": 0.0003, "grad_norm": 0.1569018198689372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43926970758, "tokens_per_second": 127397.35365565825, "data_wait_s": 1.1082996934419498e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:984f64f63a6d9c46b03440350f09af265cb985eed79cab2d9c3c24d9a17c8d2d
-size 422661
+oid sha256:2995b7adc2501c2358d806f99152239e7ba4f0537883278e739a9662edc7e176
+size 424779