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": 75611.40142043606,
-      "required_pages_per_second_upper_estimate": 0.006553804639778513
+      "required_token_positions_per_second": 73757.42736056469,
+      "required_pages_per_second_upper_estimate": 0.006393106867123158
     },
     {
       "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": 9451.425177554507,
-      "required_pages_per_second_upper_estimate": 0.003770314814725749
+      "required_token_positions_per_second": 9219.678420070586,
+      "required_pages_per_second_upper_estimate": 0.0036778675682426144
     },
     {
       "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": 18902.850355109014,
-      "required_pages_per_second_upper_estimate": 0.004372841750850612
+      "required_token_positions_per_second": 18439.356840141172,
+      "required_pages_per_second_upper_estimate": 0.004265620683370062
     },
     {
       "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": 28354.27553266352,
-      "required_pages_per_second_upper_estimate": 0.0017266266165525289
+      "required_token_positions_per_second": 27659.03526021176,
+      "required_pages_per_second_upper_estimate": 0.0016842901316040215
     },
     {
       "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": 28354.27553266352,
-      "required_pages_per_second_upper_estimate": 0.002182790620162674
+      "required_token_positions_per_second": 27659.03526021176,
+      "required_pages_per_second_upper_estimate": 0.0021292690994410865
     },
     {
       "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": 28354.27553266352,
-      "required_pages_per_second_upper_estimate": 0.3520652064599317
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27659.03526021176,
+      "required_pages_per_second_upper_estimate": 0.34343264909559285
     }
   ],
   "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": 0.12646877672699475,
-    "producer_p95_seconds": 0.3721874629991362,
-    "gpu_update_seconds": 0.9750910393795847,
-    "producer_batches_per_second": 7.9070900018166315,
-    "gpu_batches_per_second": 1.0255452666617304,
-    "producer_headroom": 7.710132608339301,
+    "producer_batch_seconds": 0.1862012017422785,
+    "producer_p95_seconds": 0.43837926800188143,
+    "gpu_update_seconds": 0.9996010251222451,
+    "producer_batches_per_second": 5.370534618697587,
+    "gpu_batches_per_second": 1.0003991341222425,
+    "producer_headroom": 5.368391910304613,
     "sustainable": true,
-    "buffer_seconds": 60.455644441534254,
+    "buffer_seconds": 60.97566253245695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314663.0149923685,
-    "upload_bytes_per_second": 5251378.38967106,
+    "generation_bytes_per_second": 321576.6813675006,
+    "upload_bytes_per_second": 5450583.341950757,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.51935803726475,
+    "estimated_full_shards_per_hour": 69.00286989945187,
     "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": 4.375384323997423,
-      "bytes": 11988381,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 0.9212507589982124,
       "bytes": 677495,
@@ -562,6 +556,12 @@
       "bytes": 5796643,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.300869512000645,
+      "bytes": 11051681,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -403,3 +403,4 @@
 {"step": 277400, "loss": 0.5671660769730806, "lr": 0.0003, "grad_norm": 0.14708343725716377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40603211126, "tokens_per_second": 148392.47630144472, "data_wait_s": 1.7530997865833342e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277500, "loss": 0.6091627143323421, "lr": 0.0003, "grad_norm": 0.15208406848093192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40617975581, "tokens_per_second": 149867.00935391747, "data_wait_s": 1.4083998394198716e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277600, "loss": 0.598962739109993, "lr": 0.0003, "grad_norm": 0.16391701987067572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40632621246, "tokens_per_second": 148583.28325976295, "data_wait_s": 2.0689993107225746e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277700, "loss": 0.5741484090685844, "lr": 0.0003, "grad_norm": 0.13775773153410134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40647240294, "tokens_per_second": 148081.28318683952, "data_wait_s": 1.3342003512661904e-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 sha256:1d22549c334208e5aee8fdfa4bede5c6f82d71a7daba674dbdab203a605bcf33
-size 864219
+oid sha256:d8090d8e41e17f51c6f39e1cd0120904db45c93640e64754ce70e98c138d3d40
+size 866337