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": 75944.73760233549,
-      "required_pages_per_second_upper_estimate": 0.006582697375192726
+      "required_token_positions_per_second": 74607.88235005584,
+      "required_pages_per_second_upper_estimate": 0.006466822150153787
     },
     {
       "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": 9493.092200291936,
-      "required_pages_per_second_upper_estimate": 0.0037869364130732153
+      "required_token_positions_per_second": 9325.98529375698,
+      "required_pages_per_second_upper_estimate": 0.003720274969585519
     },
     {
       "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": 18986.184400583872,
-      "required_pages_per_second_upper_estimate": 0.004392119615642112
+      "required_token_positions_per_second": 18651.97058751396,
+      "required_pages_per_second_upper_estimate": 0.004314805131950604
     },
     {
       "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": 28479.276600875808,
-      "required_pages_per_second_upper_estimate": 0.0017342385257767374
+      "required_token_positions_per_second": 27977.95588127094,
+      "required_pages_per_second_upper_estimate": 0.0017037107241793421
     },
     {
       "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": 28479.276600875808,
-      "required_pages_per_second_upper_estimate": 0.002192413548418759
+      "required_token_positions_per_second": 27977.95588127094,
+      "required_pages_per_second_upper_estimate": 0.002153820491679005
     },
     {
       "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": 28479.276600875808,
-      "required_pages_per_second_upper_estimate": 0.3536173013754648
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27977.95588127094,
+      "required_pages_per_second_upper_estimate": 0.3473925758504903
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 717798,
-    "producer_batch_seconds": 0.09762365248434435,
-    "producer_p95_seconds": 0.17135631800010742,
-    "gpu_update_seconds": 0.970811175700641,
-    "producer_batches_per_second": 10.243419238594534,
-    "gpu_batches_per_second": 1.030066427983066,
-    "producer_headroom": 9.944425874214526,
+    "producer_batch_seconds": 0.22681345022660793,
+    "producer_p95_seconds": 0.16623987200000556,
+    "gpu_update_seconds": 0.9882065765393597,
+    "producer_batches_per_second": 4.40890960831867,
+    "gpu_batches_per_second": 1.0119341681593945,
+    "producer_headroom": 4.356913470308082,
     "sustainable": true,
-    "buffer_seconds": 60.19029289343975,
+    "buffer_seconds": 60.280601168900944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350696.53929637535,
-    "upload_bytes_per_second": 5315003.418443578,
+    "generation_bytes_per_second": 390768.3136692248,
+    "upload_bytes_per_second": 5396015.94525555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.25131353538939,
+    "estimated_full_shards_per_hour": 83.84978349263723,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 879449,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.0847937970002022,
+      "bytes": 7859347,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 360100, "loss": 0.6022424828261137, "lr": 0.0003, "grad_norm": 0.18071048781891982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52708490102, "tokens_per_second": 127690.04584187643, "data_wait_s": 1.3590999515145086e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360200, "loss": 0.5641546994447708, "lr": 0.0003, "grad_norm": 0.1564444480186743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52723247920, "tokens_per_second": 150561.82350302662, "data_wait_s": 1.1824000466731377e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360300, "loss": 0.5836267694830894, "lr": 0.0003, "grad_norm": 0.14651163817856447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52737858357, "tokens_per_second": 148147.5032693042, "data_wait_s": 1.4541000382450875e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 360400, "loss": 0.636266840621829, "lr": 0.0003, "grad_norm": 0.17657023365254482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52752496198, "tokens_per_second": 149377.79098082113, "data_wait_s": 1.223800063598901e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca57c7d192d79a3187eb264ccb094db06c550e34f1f43dfdd67ac206d7330dd2
-size 113202
+oid sha256:7f83c0c70f426e92b7299993290e1f8b155dfe5c8b261c94fc42a5da678efb7e
+size 115320