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": 77285.57471054445,
-      "required_pages_per_second_upper_estimate": 0.005346795385715265
+      "required_token_positions_per_second": 75576.09724670161,
+      "required_pages_per_second_upper_estimate": 0.00522852976823242
     },
     {
       "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": 9660.696838818056,
-      "required_pages_per_second_upper_estimate": 0.0040447320281227
+      "required_token_positions_per_second": 9447.012155837701,
+      "required_pages_per_second_upper_estimate": 0.003955266713602424
     },
     {
       "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": 19321.39367763611,
-      "required_pages_per_second_upper_estimate": 0.004024136763635281
+      "required_token_positions_per_second": 18894.024311675403,
+      "required_pages_per_second_upper_estimate": 0.0039351269951937535
     },
     {
       "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": 28982.090516454165,
-      "required_pages_per_second_upper_estimate": 0.0017374320733715455
+      "required_token_positions_per_second": 28341.036467513102,
+      "required_pages_per_second_upper_estimate": 0.0016990018619703854
     },
     {
       "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": 28982.090516454165,
-      "required_pages_per_second_upper_estimate": 0.0022532300589104986
+      "required_token_positions_per_second": 28341.036467513102,
+      "required_pages_per_second_upper_estimate": 0.002203390926304097
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28982.090516454165,
-      "required_pages_per_second_upper_estimate": 0.37175113860076403
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28341.036467513102,
+      "required_pages_per_second_upper_estimate": 0.3635283855711587
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.131679999273274,
-    "producer_p95_seconds": 0.3379022870067274,
-    "gpu_update_seconds": 0.9539684511130502,
-    "producer_batches_per_second": 7.59416772113365,
-    "gpu_batches_per_second": 1.0482526951842508,
-    "producer_headroom": 7.24459641842259,
+    "producer_batch_seconds": 0.3686582535472098,
+    "producer_p95_seconds": 0.3421674630008056,
+    "gpu_update_seconds": 0.9755465376748829,
+    "producer_batches_per_second": 2.712539297243596,
+    "gpu_batches_per_second": 1.0250664231594728,
+    "producer_headroom": 2.64620831973305,
     "sustainable": true,
-    "buffer_seconds": 60.10001242012216,
+    "buffer_seconds": 60.483885335842736,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 587310.8773967478,
-    "upload_bytes_per_second": 5845277.591063103,
+    "generation_bytes_per_second": 533874.956817281,
+    "upload_bytes_per_second": 5968492.4491932755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.011621195921066,
+    "estimated_full_shards_per_hour": 114.55713776005575,
     "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": 1.0896003289999499,
-      "bytes": 6195866,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1199947279965272,
       "bytes": 7293697,
@@ -562,6 +556,12 @@
       "bytes": 928482,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5923971939992043,
+      "bytes": 18139508,
+      "pages": 43,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -312,3 +312,4 @@
 {"step": 516200, "loss": 0.5786738991737366, "lr": 0.0003, "grad_norm": 0.15992800576542088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75553957353, "tokens_per_second": 150329.51372906312, "data_wait_s": 1.6395999409724027e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516300, "loss": 0.6054825186729431, "lr": 0.0003, "grad_norm": 0.15628910358101797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75568475628, "tokens_per_second": 151586.38665990692, "data_wait_s": 1.6095997125376016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516400, "loss": 0.5965888500213623, "lr": 0.0003, "grad_norm": 0.1514518811000016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75583191886, "tokens_per_second": 153272.14447573066, "data_wait_s": 2.549700002418831e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 516500, "loss": 0.5942535400390625, "lr": 0.0003, "grad_norm": 0.16532523981230376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75597973591, "tokens_per_second": 153758.60466063625, "data_wait_s": 2.15519976336509e-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:df80691f0c3fe10bc839de3a8e3a150237449063571e74e5c20e04c6245333e8
-size 669915
+oid sha256:3b9a956d9c3078bde62ba918da7b0c0997d06f68a569d5b0590bdc15be9409e3
+size 672033