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": 74925.08995005884,
-      "required_pages_per_second_upper_estimate": 0.006494316900966734
+      "required_token_positions_per_second": 75499.31588946245,
+      "required_pages_per_second_upper_estimate": 0.006544089350031901
     },
     {
       "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": 9365.636243757355,
-      "required_pages_per_second_upper_estimate": 0.0037360923263752015
+      "required_token_positions_per_second": 9437.414486182806,
+      "required_pages_per_second_upper_estimate": 0.003764725740459074
     },
     {
       "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": 18731.27248751471,
-      "required_pages_per_second_upper_estimate": 0.004333150230850006
+      "required_token_positions_per_second": 18874.828972365613,
+      "required_pages_per_second_upper_estimate": 0.004366359497112419
     },
     {
       "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": 28096.908731272066,
-      "required_pages_per_second_upper_estimate": 0.0017109543286470395
+      "required_token_positions_per_second": 28312.243458548415,
+      "required_pages_per_second_upper_estimate": 0.0017240670837641687
     },
     {
       "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": 28096.908731272066,
-      "required_pages_per_second_upper_estimate": 0.0021629778113546553
+      "required_token_positions_per_second": 28312.243458548415,
+      "required_pages_per_second_upper_estimate": 0.0021795548747450652
     },
     {
       "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": 28096.908731272066,
-      "required_pages_per_second_upper_estimate": 0.348869572137925
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28312.243458548415,
+      "required_pages_per_second_upper_estimate": 0.35154330877172496
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.6089515988203402,
-    "producer_p95_seconds": 0.23256230900005903,
-    "gpu_update_seconds": 0.9840228426705024,
-    "producer_batches_per_second": 1.642166638427747,
-    "gpu_batches_per_second": 1.0162365715882549,
-    "producer_headroom": 1.6159294836843345,
+    "producer_batch_seconds": 0.060175617078101595,
+    "producer_p95_seconds": 0.19274111300001096,
+    "gpu_update_seconds": 0.9765386498063664,
+    "producer_batches_per_second": 16.618026512334815,
+    "gpu_batches_per_second": 1.024025009351433,
+    "producer_headroom": 16.22814517280184,
     "sustainable": true,
-    "buffer_seconds": 60.02539340290065,
+    "buffer_seconds": 60.54539628799472,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251067.05647301415,
+    "generation_bytes_per_second": 262208.8815222158,
     "upload_bytes_per_second": 5815948.363474923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.873145777157006,
+    "estimated_full_shards_per_hour": 56.26392206430297,
     "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": [

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

@@ -120,3 +120,4 @@
 {"step": 367000, "loss": 0.5593036767095327, "lr": 0.0003, "grad_norm": 0.17470027523121276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53718637757, "tokens_per_second": 149394.50868001793, "data_wait_s": 1.003299985313788e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 367100, "loss": 0.5900400169193745, "lr": 0.0003, "grad_norm": 0.16538984639627727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53733206556, "tokens_per_second": 125406.44772599245, "data_wait_s": 1.0171001122216694e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 367200, "loss": 0.5813782624900341, "lr": 0.0003, "grad_norm": 0.16785935678550754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53747982092, "tokens_per_second": 150955.13384048344, "data_wait_s": 1.3026998203713447e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 367300, "loss": 0.6027555800974369, "lr": 0.0003, "grad_norm": 0.17217006902609083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53762635012, "tokens_per_second": 149822.0558716968, "data_wait_s": 1.3308999768923968e-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 sha256:2a9b3ee3ef034c2854be6091c663f1c58b376996eaf6446a0dede44bfe6673b6
-size 260127
+oid sha256:e17cd6cf0b8efedead4ef5d6703d920f540131d97bf73469fa1a7ead5dce0c2c
+size 262245