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": 73846.08641265031,
-      "required_pages_per_second_upper_estimate": 0.006400791609053627
+      "required_token_positions_per_second": 74985.8851158751,
+      "required_pages_per_second_upper_estimate": 0.00649958647185575
     },
     {
       "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": 9230.76080158129,
-      "required_pages_per_second_upper_estimate": 0.0036822884959236035
+      "required_token_positions_per_second": 9373.235639484388,
+      "required_pages_per_second_upper_estimate": 0.0037391238389518065
     },
     {
       "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": 18461.52160316258,
-      "required_pages_per_second_upper_estimate": 0.004270748111208558
+      "required_token_positions_per_second": 18746.471278968776,
+      "required_pages_per_second_upper_estimate": 0.00433666620376331
     },
     {
       "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": 27692.282404743866,
-      "required_pages_per_second_upper_estimate": 0.0016863147082717403
+      "required_token_positions_per_second": 28119.706918453165,
+      "required_pages_per_second_upper_estimate": 0.0017123426186335246
     },
     {
       "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": 27692.282404743866,
-      "required_pages_per_second_upper_estimate": 0.0021318285566611548
+      "required_token_positions_per_second": 28119.706918453165,
+      "required_pages_per_second_upper_estimate": 0.0021647328789132776
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27692.282404743866,
-      "required_pages_per_second_upper_estimate": 0.3438454673596467
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28119.706918453165,
+      "required_pages_per_second_upper_estimate": 0.34915264932209
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09940460011711139,
-    "producer_p95_seconds": 0.40154402799817035,
-    "gpu_update_seconds": 0.9984009117018003,
-    "producer_batches_per_second": 10.059896612650437,
-    "gpu_batches_per_second": 1.0016016494771363,
-    "producer_headroom": 10.04380994969605,
+    "producer_batch_seconds": 0.5042326117184075,
+    "producer_p95_seconds": 0.4362888989999192,
+    "gpu_update_seconds": 0.9832250414337137,
+    "producer_batches_per_second": 1.9832116700901876,
+    "gpu_batches_per_second": 1.0170611587982192,
+    "producer_headroom": 1.9499433764962493,
     "sustainable": true,
-    "buffer_seconds": 60.90245561380982,
+    "buffer_seconds": 60.95995256889025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351732.5800916723,
-    "upload_bytes_per_second": 5352284.678813212,
+    "generation_bytes_per_second": 322271.4019048588,
+    "upload_bytes_per_second": 5519370.637359089,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.47362377226474,
+    "estimated_full_shards_per_hour": 69.15194075450252,
     "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": 1.9025495240020973,
-      "bytes": 2606249,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0264649749988166,
       "bytes": 10585305,
@@ -562,6 +556,12 @@
       "bytes": 1166291,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4934369809998316,
+      "bytes": 13370474,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 354100, "loss": 0.5691184941679239, "lr": 0.0003, "grad_norm": 0.1687424969750012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51830503624, "tokens_per_second": 149911.55515781548, "data_wait_s": 1.4862001989968121e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354200, "loss": 0.5616220980882645, "lr": 0.0003, "grad_norm": 0.13573110957447848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51845169975, "tokens_per_second": 148393.61423746162, "data_wait_s": 1.0702999134082347e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354300, "loss": 0.5982334297150373, "lr": 0.0003, "grad_norm": 0.15875207837742716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51859864246, "tokens_per_second": 149674.6180730222, "data_wait_s": 2.284299989696592e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 354400, "loss": 0.5938422046601772, "lr": 0.0003, "grad_norm": 0.16056053732413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51874410027, "tokens_per_second": 148326.8045222818, "data_wait_s": 1.3731994840782136e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f864768433614f07392ec6a1d373e6a37020933418188164e3f5529d34ec91b
-size 838020
+oid sha256:17c03f6dbc72bab741a5b05f79761bba575166e257df4c05cdb29ba209a0d628
+size 840138