Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77047.17453922032,
-      "required_pages_per_second_upper_estimate": 0.005330302308698453
+      "required_token_positions_per_second": 76545.33439431181,
+      "required_pages_per_second_upper_estimate": 0.005295583843044119
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9630.89681740254,
-      "required_pages_per_second_upper_estimate": 0.004032255381451234
+      "required_token_positions_per_second": 9568.166799288976,
+      "required_pages_per_second_upper_estimate": 0.004005991632818823
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19261.79363480508,
-      "required_pages_per_second_upper_estimate": 0.0040117236464724025
+      "required_token_positions_per_second": 19136.333598577952,
+      "required_pages_per_second_upper_estimate": 0.003985593629529941
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28892.690452207622,
-      "required_pages_per_second_upper_estimate": 0.0017320726760259595
+      "required_token_positions_per_second": 28704.500397866927,
+      "required_pages_per_second_upper_estimate": 0.0017207909696178097
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28892.690452207622,
-      "required_pages_per_second_upper_estimate": 0.0022462795971446544
+      "required_token_positions_per_second": 28704.500397866927,
+      "required_pages_per_second_upper_estimate": 0.002231648648180234
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28892.690452207622,
-      "required_pages_per_second_upper_estimate": 0.3706044105669196
+      "required_token_positions_per_second": 28704.500397866927,
+      "required_pages_per_second_upper_estimate": 0.3681905106125746
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12787330923441687,
-    "producer_p95_seconds": 0.2917162769990682,
-    "gpu_update_seconds": 0.9569202302476294,
-    "producer_batches_per_second": 7.82024025175421,
-    "gpu_batches_per_second": 1.0450191859160742,
-    "producer_headroom": 7.483346102300417,
+    "producer_batch_seconds": 0.15295083794521247,
+    "producer_p95_seconds": 0.27993752299880725,
+    "gpu_update_seconds": 0.9631939109469595,
+    "producer_batches_per_second": 6.538048522219953,
+    "gpu_batches_per_second": 1.0382125433256266,
+    "producer_headroom": 6.297408526078026,
     "sustainable": true,
-    "buffer_seconds": 60.28597450560065,
+    "buffer_seconds": 60.681216389658445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289121.9005553929,
-    "upload_bytes_per_second": 6065819.655259725,
+    "generation_bytes_per_second": 320675.116567759,
+    "upload_bytes_per_second": 6157891.941618039,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.038829445804026,
+    "estimated_full_shards_per_hour": 68.80941508078173,
     "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": 1.182365384996956,
-      "bytes": 6177943,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8839036259996647,
       "bytes": 2483088,
@@ -562,6 +556,12 @@
       "bytes": 861201,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3237921879990608,
+      "bytes": 13505764,
+      "pages": 42,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -267,3 +267,4 @@
 {"step": 511700, "loss": 0.5576717257499695, "lr": 0.0003, "grad_norm": 0.17270685077586168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74895815676, "tokens_per_second": 151313.98547595218, "data_wait_s": 2.2541000362252817e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511800, "loss": 0.5815263986587524, "lr": 0.0003, "grad_norm": 0.14792045336126503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74910379444, "tokens_per_second": 151404.08318931167, "data_wait_s": 1.487600093241781e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511900, "loss": 0.6044458150863647, "lr": 0.0003, "grad_norm": 0.15826686798138512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74925019221, "tokens_per_second": 151923.77474090797, "data_wait_s": 1.386900112265721e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 512000, "loss": 0.5737632513046265, "lr": 0.0003, "grad_norm": 0.17436264043066957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74939702847, "tokens_per_second": 152854.79894645113, "data_wait_s": 1.6078000044217333e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:366331383a49e026a4330e22d8aaaa472249dc2c9ce572702c5f52c9691e3648
-size 573822
+oid sha256:517875e1f5b5f397dba711e53bc7d71022878b1bc9f35703e835de139b256b6c
+size 575940