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": 76329.52372956047,
-      "required_pages_per_second_upper_estimate": 0.006616049661469065
+      "required_token_positions_per_second": 75877.73819165355,
+      "required_pages_per_second_upper_estimate": 0.00657689003608327
     },
     {
       "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": 9541.19046619506,
-      "required_pages_per_second_upper_estimate": 0.0038061235304751315
+      "required_token_positions_per_second": 9484.717273956694,
+      "required_pages_per_second_upper_estimate": 0.00378359552974178
     },
     {
       "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": 19082.38093239012,
-      "required_pages_per_second_upper_estimate": 0.004414372990274351
+      "required_token_positions_per_second": 18969.434547913388,
+      "required_pages_per_second_upper_estimate": 0.004388244832014134
     },
     {
       "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": 28623.571398585176,
-      "required_pages_per_second_upper_estimate": 0.0017430253219009428
+      "required_token_positions_per_second": 28454.151821870084,
+      "required_pages_per_second_upper_estimate": 0.0017327085585548167
     },
     {
       "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": 28623.571398585176,
-      "required_pages_per_second_upper_estimate": 0.002203521761379991
+      "required_token_positions_per_second": 28454.151821870084,
+      "required_pages_per_second_upper_estimate": 0.0021904793733810587
     },
     {
       "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": 28623.571398585176,
-      "required_pages_per_second_upper_estimate": 0.3554089598393928
+      "required_token_positions_per_second": 28454.151821870084,
+      "required_pages_per_second_upper_estimate": 0.3533053357074399
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.16264687696889268,
-    "producer_p95_seconds": 0.3858460460032802,
-    "gpu_update_seconds": 0.9659172021198795,
-    "producer_batches_per_second": 6.148288972011782,
-    "gpu_batches_per_second": 1.0352854238492901,
-    "producer_headroom": 5.938738081670131,
+    "producer_batch_seconds": 0.11995669424209154,
+    "producer_p95_seconds": 0.3888764750008704,
+    "gpu_update_seconds": 0.9716683938809074,
+    "producer_batches_per_second": 8.336341763318703,
+    "gpu_batches_per_second": 1.0291576903164816,
+    "producer_headroom": 8.100159812006217,
     "sustainable": true,
-    "buffer_seconds": 60.85278373355241,
+    "buffer_seconds": 60.24344042061626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282437.18975245336,
-    "upload_bytes_per_second": 5461863.241892408,
+    "generation_bytes_per_second": 282652.15087444184,
+    "upload_bytes_per_second": 5437175.676486521,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.60444611959649,
+    "estimated_full_shards_per_hour": 60.6505717723364,
     "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": 1.272736505001376,
-      "bytes": 7611348,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8344998959983059,
       "bytes": 1093697,
@@ -562,6 +556,12 @@
       "bytes": 5253625,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.5616185160033638,
+      "bytes": 7297224,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 150500, "loss": 0.6060477271676064, "lr": 0.0003, "grad_norm": 0.13552522216887578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22025162642, "tokens_per_second": 148150.53943653347, "data_wait_s": 8.285998774226755e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 150600, "loss": 0.6056232452392578, "lr": 0.0003, "grad_norm": 0.13553303769077266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22039723020, "tokens_per_second": 149647.2460471847, "data_wait_s": 1.2923002941533923e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150700, "loss": 0.5843577682971954, "lr": 0.0003, "grad_norm": 0.14000978061212047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22054300582, "tokens_per_second": 149173.14354409723, "data_wait_s": 1.2389995390549302e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150800, "loss": 0.5710970014333725, "lr": 0.0003, "grad_norm": 0.14444581917404686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22069015443, "tokens_per_second": 150694.64432425157, "data_wait_s": 1.0910996934399009e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fa09c3c5fdd20c7111dbcec040c28d41f6b5555c499abebc27bb2c08b63d01c
-size 761772
+oid sha256:fe4d6043b894063640b11426b045137dd86c407fa521337d20859651a53a98a7
+size 763890