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": 77066.38715249821,
-      "required_pages_per_second_upper_estimate": 0.005331631481864422
+      "required_token_positions_per_second": 77059.36635315721,
+      "required_pages_per_second_upper_estimate": 0.005331145766675514
     },
     {
       "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": 9633.298394062276,
-      "required_pages_per_second_upper_estimate": 0.004033260871448042
+      "required_token_positions_per_second": 9632.420794144651,
+      "required_pages_per_second_upper_estimate": 0.004032893438688903
     },
     {
       "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": 19266.596788124552,
-      "required_pages_per_second_upper_estimate": 0.004012724016641178
+      "required_token_positions_per_second": 19264.841588289302,
+      "required_pages_per_second_upper_estimate": 0.004012358454803228
     },
     {
       "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": 28899.89518218683,
-      "required_pages_per_second_upper_estimate": 0.0017325045885872273
+      "required_token_positions_per_second": 28897.26238243395,
+      "required_pages_per_second_upper_estimate": 0.0017323467562620947
     },
     {
       "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": 28899.89518218683,
-      "required_pages_per_second_upper_estimate": 0.00224683973321028
+      "required_token_positions_per_second": 28897.26238243395,
+      "required_pages_per_second_upper_estimate": 0.002246635044609956
     },
     {
       "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": 28899.89518218683,
-      "required_pages_per_second_upper_estimate": 0.37069682510725654
+      "required_token_positions_per_second": 28897.26238243395,
+      "required_pages_per_second_upper_estimate": 0.37066305437890673
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04997484998437329,
-    "producer_p95_seconds": 0.16840392499943846,
-    "gpu_update_seconds": 0.9566816704941385,
-    "producer_batches_per_second": 20.010065068983526,
-    "gpu_batches_per_second": 1.0452797736612711,
-    "producer_headroom": 19.14326247689157,
+    "producer_batch_seconds": 0.32835209026564627,
+    "producer_p95_seconds": 0.20609207799952856,
+    "gpu_update_seconds": 0.9567688327738149,
+    "producer_batches_per_second": 3.045511296093688,
+    "gpu_batches_per_second": 1.045184547975765,
+    "producer_headroom": 2.913850287963026,
     "sustainable": true,
-    "buffer_seconds": 60.270945241130725,
+    "buffer_seconds": 60.27643646475033,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384821.4852630652,
-    "upload_bytes_per_second": 2833567.300453067,
+    "generation_bytes_per_second": 376946.3065905582,
+    "upload_bytes_per_second": 2862600.0249659023,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.5737325517556,
+    "estimated_full_shards_per_hour": 80.88390253341255,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 6356080,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.188576070999261,
+      "bytes": 5752447,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 652500, "loss": 0.5530551075935364, "lr": 0.0001, "grad_norm": 0.1764890985252482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95508158789, "tokens_per_second": 152569.6998663404, "data_wait_s": 9.280999620386865e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 652600, "loss": 0.5564659237861633, "lr": 0.0001, "grad_norm": 0.18541973571262774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95522942862, "tokens_per_second": 154391.4041044394, "data_wait_s": 8.222999895224348e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 652700, "loss": 0.5584211349487305, "lr": 0.0001, "grad_norm": 0.1893255976917713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95537732030, "tokens_per_second": 154305.00237994987, "data_wait_s": 9.487999705015682e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 652800, "loss": 0.5506278872489929, "lr": 0.0001, "grad_norm": 0.21742603238567818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95552335978, "tokens_per_second": 152711.57912437886, "data_wait_s": 1.398899985360913e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:939520d4967a1da71da5c75d01d88d249df6ca7f554e4d6b0e2d13787461f51c
-size 164034
+oid sha256:47ce11ba306c5336c5c03a3abbf5a6e2ac215bbd51bb2254f60ab801d1135741
+size 166152