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": 75711.9398713743,
-      "required_pages_per_second_upper_estimate": 0.005237927676205592
+      "required_token_positions_per_second": 76892.09570848692,
+      "required_pages_per_second_upper_estimate": 0.005319573595355854
     },
     {
       "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": 9463.992483921787,
-      "required_pages_per_second_upper_estimate": 0.003962376022381659
+      "required_token_positions_per_second": 9611.511963560864,
+      "required_pages_per_second_upper_estimate": 0.004024139347949504
     },
     {
       "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": 18927.984967843575,
-      "required_pages_per_second_upper_estimate": 0.003942200104270854
+      "required_token_positions_per_second": 19223.02392712173,
+      "required_pages_per_second_upper_estimate": 0.004003648938787908
     },
     {
       "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": 28391.977451765364,
-      "required_pages_per_second_upper_estimate": 0.0017020556961939292
+      "required_token_positions_per_second": 28834.53589068259,
+      "required_pages_per_second_upper_estimate": 0.0017285863988594078
     },
     {
       "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": 28391.977451765364,
-      "required_pages_per_second_upper_estimate": 0.002207351363764001
+      "required_token_positions_per_second": 28834.53589068259,
+      "required_pages_per_second_upper_estimate": 0.0022417583357809663
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28391.977451765364,
-      "required_pages_per_second_upper_estimate": 0.36418180182097926
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28834.53589068259,
+      "required_pages_per_second_upper_estimate": 0.3698584662931798
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07457380725014673,
-    "producer_p95_seconds": 0.2536876190024486,
-    "gpu_update_seconds": 0.9737962086991195,
-    "producer_batches_per_second": 13.409533948637609,
-    "gpu_batches_per_second": 1.0269089066755412,
-    "producer_headroom": 13.058153319605438,
+    "producer_batch_seconds": 0.4236891052736098,
+    "producer_p95_seconds": 0.27590801600308623,
+    "gpu_update_seconds": 0.958850182462413,
+    "producer_batches_per_second": 2.3602211799952237,
+    "gpu_batches_per_second": 1.0429157946572118,
+    "producer_headroom": 2.263098509090072,
     "sustainable": true,
-    "buffer_seconds": 60.37536493934541,
+    "buffer_seconds": 60.40756149513202,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381443.12327331054,
-    "upload_bytes_per_second": 5911135.943370028,
+    "generation_bytes_per_second": 365117.29735174275,
+    "upload_bytes_per_second": 6000144.907552854,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.84881471299636,
+    "estimated_full_shards_per_hour": 78.34567251600467,
     "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": 1.176046257998678,
-      "bytes": 6195296,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9275896610015479,
       "bytes": 1143867,
@@ -562,6 +556,12 @@
       "bytes": 6158826,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2382891359993664,
+      "bytes": 13055404,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -237,3 +237,4 @@
 {"step": 508700, "loss": 0.5609819889068604, "lr": 0.0003, "grad_norm": 0.14620061298802478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74456476855, "tokens_per_second": 153012.41598693808, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 508800, "loss": 0.5915711522102356, "lr": 0.0003, "grad_norm": 0.16639111225339134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74470981055, "tokens_per_second": 150459.7853398257, "data_wait_s": 2.034299905062653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 508900, "loss": 0.6434043049812317, "lr": 0.0003, "grad_norm": 0.17744187198432396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74485676663, "tokens_per_second": 153030.7579752688, "data_wait_s": 1.1971002095378935e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 509000, "loss": 0.6046046018600464, "lr": 0.0003, "grad_norm": 0.1796017689311492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74500347953, "tokens_per_second": 152458.45778727342, "data_wait_s": 8.307997632073238e-06, "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 sha256:33b3820d3caadc4cc036f39203256168b54174b9b61bb9367074308bfd977411
-size 509499
+oid sha256:7c448b1b041bfdcbbfcb3f54661cedaeb68ffe09a7997cc68a410e2113e9c4a8
+size 511617