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": 76943.11672433751,
-      "required_pages_per_second_upper_estimate": 0.005323103347617465
+      "required_token_positions_per_second": 77290.30649914793,
+      "required_pages_per_second_upper_estimate": 0.005347122741830117
     },
     {
       "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": 9617.889590542189,
-      "required_pages_per_second_upper_estimate": 0.004026809527186589
+      "required_token_positions_per_second": 9661.288312393492,
+      "required_pages_per_second_upper_estimate": 0.004044979665757362
     },
     {
       "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": 19235.779181084377,
-      "required_pages_per_second_upper_estimate": 0.00400630552180966
+      "required_token_positions_per_second": 19322.576624786983,
+      "required_pages_per_second_upper_estimate": 0.0040243831403303934
     },
     {
       "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": 28853.668771626566,
-      "required_pages_per_second_upper_estimate": 0.0017297333858577815
+      "required_token_positions_per_second": 28983.864937180475,
+      "required_pages_per_second_upper_estimate": 0.0017375384471847043
     },
     {
       "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": 28853.668771626566,
-      "required_pages_per_second_upper_estimate": 0.002243245833117711
+      "required_token_positions_per_second": 28983.864937180475,
+      "required_pages_per_second_upper_estimate": 0.002253368012317117
     },
     {
       "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": 28853.668771626566,
-      "required_pages_per_second_upper_estimate": 0.3701038823466421
+      "required_token_positions_per_second": 28983.864937180475,
+      "required_pages_per_second_upper_estimate": 0.3717738989646166
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1178831686873707,
-    "producer_p95_seconds": 0.26381924000088475,
-    "gpu_update_seconds": 0.9582143684683812,
-    "producer_batches_per_second": 8.482975229924694,
-    "gpu_batches_per_second": 1.0436078114737617,
-    "producer_headroom": 8.128508752675211,
+    "producer_batch_seconds": 0.18497176878898358,
+    "producer_p95_seconds": 0.2531597890010744,
+    "gpu_update_seconds": 0.9539100482259414,
+    "producer_batches_per_second": 5.406230402331306,
+    "gpu_batches_per_second": 1.0483168741746411,
+    "producer_headroom": 5.1570575038084066,
     "sustainable": true,
-    "buffer_seconds": 60.36750521350801,
+    "buffer_seconds": 60.096333038234306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 593935.146880332,
-    "upload_bytes_per_second": 5021034.13757648,
+    "generation_bytes_per_second": 566989.6717913321,
+    "upload_bytes_per_second": 4948956.679366621,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.72232820895896,
+    "estimated_full_shards_per_hour": 60.8313923611878,
     "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": 1.5480925499996374,
-      "bytes": 11721285,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.0147747470000468,
       "bytes": 1162064,
@@ -562,6 +556,12 @@
       "bytes": 13448331,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.4291834129980998,
+      "bytes": 4870018,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 906700, "loss": 0.5318596363067627, "lr": 3e-05, "grad_norm": 0.3319238812255569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132714852166, "tokens_per_second": 150475.67974385415, "data_wait_s": 1.4097000530455261e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906800, "loss": 0.5484407544136047, "lr": 3e-05, "grad_norm": 0.3447702822185023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132729456073, "tokens_per_second": 151491.73993199418, "data_wait_s": 1.410899858456105e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906900, "loss": 0.5580467581748962, "lr": 3e-05, "grad_norm": 0.3092025897628717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132744019053, "tokens_per_second": 151957.30139648938, "data_wait_s": 1.499800055171363e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 907000, "loss": 0.5452108979225159, "lr": 3e-05, "grad_norm": 0.32432525863845285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132758583743, "tokens_per_second": 151527.62240822273, "data_wait_s": 1.5677996998419985e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be9e0488eaddab6cc7d9458073ef55ab7e313f5f7cdaf944f3c30a2909859aad
-size 467139
+oid sha256:12fb41a921e814016f9551bb1d4b7e3c9710d20e580841b4c86912096d39b5ef
+size 469257