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": 74778.77638337637,
-      "required_pages_per_second_upper_estimate": 0.006481634811835043
+      "required_token_positions_per_second": 75419.06243140057,
+      "required_pages_per_second_upper_estimate": 0.006537133183687627
     },
     {
       "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": 9347.347047922047,
-      "required_pages_per_second_upper_estimate": 0.003728796492708651
+      "required_token_positions_per_second": 9427.382803925071,
+      "required_pages_per_second_upper_estimate": 0.0037607239524194477
     },
     {
       "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": 18694.694095844094,
-      "required_pages_per_second_upper_estimate": 0.004324688463694715
+      "required_token_positions_per_second": 18854.765607850142,
+      "required_pages_per_second_upper_estimate": 0.0043617181908349205
     },
     {
       "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": 28042.04114376614,
-      "required_pages_per_second_upper_estimate": 0.0017076131804359125
+      "required_token_positions_per_second": 28282.14841177521,
+      "required_pages_per_second_upper_estimate": 0.0017222344533121881
     },
     {
       "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": 28042.04114376614,
-      "required_pages_per_second_upper_estimate": 0.002158753952585246
+      "required_token_positions_per_second": 28282.14841177521,
+      "required_pages_per_second_upper_estimate": 0.002177238074736042
     },
     {
       "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": 28042.04114376614,
-      "required_pages_per_second_upper_estimate": 0.3481883003311042
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28282.14841177521,
+      "required_pages_per_second_upper_estimate": 0.35116962901244414
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.23463934814839504,
-    "producer_p95_seconds": 0.17552479299956758,
-    "gpu_update_seconds": 0.985948200355817,
-    "producer_batches_per_second": 4.261859777105932,
-    "gpu_batches_per_second": 1.0142520668318193,
-    "producer_headroom": 4.201972977406436,
+    "producer_batch_seconds": 0.049299361179667045,
+    "producer_p95_seconds": 0.16747984799985716,
+    "gpu_update_seconds": 0.9775777850203492,
+    "producer_batches_per_second": 20.284238498661086,
+    "gpu_batches_per_second": 1.0229365021620085,
+    "producer_headroom": 19.829420942345596,
     "sustainable": true,
-    "buffer_seconds": 60.14284022170484,
+    "buffer_seconds": 60.609822671261654,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 564117.4594634526,
-    "upload_bytes_per_second": 4560660.040824975,
+    "generation_bytes_per_second": 564038.7175145241,
+    "upload_bytes_per_second": 4633760.506625408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.52323740924684,
+    "estimated_full_shards_per_hour": 60.5147893146362,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 8195692,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.6225744519997534,
+      "bytes": 9437094,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 239400, "loss": 0.6398460641503334, "lr": 0.0003, "grad_norm": 0.17507099564559192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35040837015, "tokens_per_second": 148336.3790418062, "data_wait_s": 1.5294999684556387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239500, "loss": 0.6293351035565138, "lr": 0.0003, "grad_norm": 0.14969589073221598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35055559803, "tokens_per_second": 150727.93092820636, "data_wait_s": 1.1714999800460646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239600, "loss": 0.54215855197981, "lr": 0.0003, "grad_norm": 0.16648286641590115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35070155866, "tokens_per_second": 147424.67963321402, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 239700, "loss": 0.6767937075346708, "lr": 0.0003, "grad_norm": 0.14570195230763597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35084637670, "tokens_per_second": 147028.0926855148, "data_wait_s": 1.6646999938529916e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbf1b28109e247cc5e801ea5a9efdaeb4daf8271b350f2aa5884aecec1f2ea11
-size 53115
+oid sha256:abe923fa86712746cc74d992bb579ddc3f79abfb8689e28b4211583e2026ab9f
+size 55233