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": 77108.51888582605,
-      "required_pages_per_second_upper_estimate": 0.005334546252935133
+      "required_token_positions_per_second": 77181.09950879787,
+      "required_pages_per_second_upper_estimate": 0.00533956754884257
     },
     {
       "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": 9638.564860728256,
-      "required_pages_per_second_upper_estimate": 0.004035465831064758
+      "required_token_positions_per_second": 9647.637438599733,
+      "required_pages_per_second_upper_estimate": 0.004039264329962575
     },
     {
       "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": 19277.129721456513,
-      "required_pages_per_second_upper_estimate": 0.004014917748882094
+      "required_token_positions_per_second": 19295.274877199467,
+      "required_pages_per_second_upper_estimate": 0.0040186969063034155
     },
     {
       "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": 28915.69458218477,
-      "required_pages_per_second_upper_estimate": 0.0017334517384928168
+      "required_token_positions_per_second": 28942.9123157992,
+      "required_pages_per_second_upper_estimate": 0.0017350834000639291
     },
     {
       "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": 28915.69458218477,
-      "required_pages_per_second_upper_estimate": 0.0022480680670659034
+      "required_token_positions_per_second": 28942.9123157992,
+      "required_pages_per_second_upper_estimate": 0.0022501841261362674
     },
     {
       "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": 28915.69458218477,
-      "required_pages_per_second_upper_estimate": 0.3708994828463561
+      "required_token_positions_per_second": 28942.9123157992,
+      "required_pages_per_second_upper_estimate": 0.3712486027090366
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11619011374250476,
-    "producer_p95_seconds": 0.34990405599819496,
-    "gpu_update_seconds": 0.9561589441131458,
-    "producer_batches_per_second": 8.606584224679859,
-    "gpu_batches_per_second": 1.0458512218672154,
-    "producer_headroom": 8.229262484690752,
+    "producer_batch_seconds": 0.16873864537512873,
+    "producer_p95_seconds": 0.36738264599989634,
+    "gpu_update_seconds": 0.9552597782258304,
+    "producer_batches_per_second": 5.926324688555282,
+    "gpu_batches_per_second": 1.0468356595702837,
+    "producer_headroom": 5.661179607683582,
     "sustainable": true,
-    "buffer_seconds": 60.238013479128185,
+    "buffer_seconds": 60.181366028227316,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366835.02425556857,
+    "generation_bytes_per_second": 356533.5851442142,
     "upload_bytes_per_second": 4909669.824765278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.71425672292989,
+    "estimated_full_shards_per_hour": 76.50380769486256,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 596200, "loss": 0.5732657313346863, "lr": 0.0003, "grad_norm": 0.18199784911604647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87265546722, "tokens_per_second": 150874.52269238717, "data_wait_s": 1.0335003025829792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596300, "loss": 0.5600859522819519, "lr": 0.0003, "grad_norm": 0.1622003101067814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87280233172, "tokens_per_second": 153579.91889956864, "data_wait_s": 1.4345998351927847e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596400, "loss": 0.6058626174926758, "lr": 0.0003, "grad_norm": 0.15734993949551213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87295058543, "tokens_per_second": 153126.0134532248, "data_wait_s": 1.3318000128492713e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 596500, "loss": 0.5947234034538269, "lr": 0.0003, "grad_norm": 0.1938059087422956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87309746644, "tokens_per_second": 152386.78016610106, "data_wait_s": 8.946997695602477e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76dcbac308ddb5e2572bdf73ad5bee20fd2b5d945880de3d4147a918f54f3602
-size 838803
+oid sha256:2fa02a04d6020705ac93bdb03fee6e0d3bc9407ccc88288824cd4a25a4a02d20
+size 840921