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": 76952.6696927135,
-      "required_pages_per_second_upper_estimate": 0.005323764244135657
+      "required_token_positions_per_second": 75274.3345656096,
+      "required_pages_per_second_upper_estimate": 0.005207653125768589
     },
     {
       "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": 9619.083711589188,
-      "required_pages_per_second_upper_estimate": 0.004027309480732885
+      "required_token_positions_per_second": 9409.2918207012,
+      "required_pages_per_second_upper_estimate": 0.0039394739969709405
     },
     {
       "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": 19238.167423178376,
-      "required_pages_per_second_upper_estimate": 0.00400680292965566
+      "required_token_positions_per_second": 18818.5836414024,
+      "required_pages_per_second_upper_estimate": 0.003919414693080151
     },
     {
       "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": 28857.251134767565,
-      "required_pages_per_second_upper_estimate": 0.0017299481430581326
+      "required_token_positions_per_second": 28227.8754621036,
+      "required_pages_per_second_upper_estimate": 0.0016922180324829362
     },
     {
       "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": 28857.251134767565,
-      "required_pages_per_second_upper_estimate": 0.0022435243460947733
+      "required_token_positions_per_second": 28227.8754621036,
+      "required_pages_per_second_upper_estimate": 0.0021945931558761678
     },
     {
       "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": 28857.251134767565,
-      "required_pages_per_second_upper_estimate": 0.370149833054573
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28227.8754621036,
+      "required_pages_per_second_upper_estimate": 0.36207687769658675
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11958894939073161,
-    "producer_p95_seconds": 0.36296374799712794,
-    "gpu_update_seconds": 0.9580954149402455,
-    "producer_batches_per_second": 8.361976629903415,
-    "gpu_batches_per_second": 1.0437373818998685,
-    "producer_headroom": 8.011571468947949,
+    "producer_batch_seconds": 0.1684293638828649,
+    "producer_p95_seconds": 0.3530404089979129,
+    "gpu_update_seconds": 0.9794573465905327,
+    "producer_batches_per_second": 5.937207010384813,
+    "gpu_batches_per_second": 1.0209735048503905,
+    "producer_headroom": 5.815241024550218,
     "sustainable": true,
-    "buffer_seconds": 60.36001114123547,
+    "buffer_seconds": 60.72635548861303,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371868.8285592774,
-    "upload_bytes_per_second": 4589272.036223166,
+    "generation_bytes_per_second": 383167.50969957816,
+    "upload_bytes_per_second": 4561313.382369415,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.79439394553891,
+    "estimated_full_shards_per_hour": 82.21882789841183,
     "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": 1.3582085800007917,
-      "bytes": 8243145,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.29941092199806,
       "bytes": 6125022,
@@ -562,6 +556,12 @@
       "bytes": 5260953,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3634449710007175,
+      "bytes": 5626486,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 644300, "loss": 0.5643414855003357, "lr": 0.0001, "grad_norm": 0.1783961950379836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94308908391, "tokens_per_second": 154237.19646603084, "data_wait_s": 1.6951002180576324e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644400, "loss": 0.556706964969635, "lr": 0.0001, "grad_norm": 0.17133278942983368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94323676420, "tokens_per_second": 153007.10796972917, "data_wait_s": 1.392999547533691e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644500, "loss": 0.564680278301239, "lr": 0.0001, "grad_norm": 0.19166007721770675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94338417294, "tokens_per_second": 152386.79696061014, "data_wait_s": 1.418799365637824e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 644600, "loss": 0.5822890400886536, "lr": 0.0001, "grad_norm": 0.22338748039960604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94352959361, "tokens_per_second": 150792.3327680029, "data_wait_s": 1.3247001334093511e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c1b84ea6a095bc0d50423eb4d1182e9bc34ec88bbcf2753d2adee8c96a49910
-size 855747
+oid sha256:fa02e693cfb4ef8bfe65f5150cd6ab38b11949e3d73135131c208741a64c6c9d
+size 857865