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": 77475.01838845987,
-      "required_pages_per_second_upper_estimate": 0.005359901538923348
+      "required_token_positions_per_second": 77410.26084293822,
+      "required_pages_per_second_upper_estimate": 0.005355421461665954
     },
     {
       "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": 9684.377298557483,
-      "required_pages_per_second_upper_estimate": 0.0040546465421113665
+      "required_token_positions_per_second": 9676.282605367278,
+      "required_pages_per_second_upper_estimate": 0.004051257463109043
     },
     {
       "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": 19368.754597114967,
-      "required_pages_per_second_upper_estimate": 0.004034000794171289
+      "required_token_positions_per_second": 19352.565210734556,
+      "required_pages_per_second_upper_estimate": 0.004030628971931079
     },
     {
       "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": 29053.13189567245,
-      "required_pages_per_second_upper_estimate": 0.0017416908955817764
+      "required_token_positions_per_second": 29028.847816101832,
+      "required_pages_per_second_upper_estimate": 0.0017402351020911605
     },
     {
       "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": 29053.13189567245,
-      "required_pages_per_second_upper_estimate": 0.0022587532136667168
+      "required_token_positions_per_second": 29028.847816101832,
+      "required_pages_per_second_upper_estimate": 0.0022568652332944786
     },
     {
       "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": 29053.13189567245,
-      "required_pages_per_second_upper_estimate": 0.37266238113508615
+      "required_token_positions_per_second": 29028.847816101832,
+      "required_pages_per_second_upper_estimate": 0.37235089103656743
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08723904333591648,
-    "producer_p95_seconds": 0.15371741600029054,
-    "gpu_update_seconds": 0.9516357857487389,
-    "producer_batches_per_second": 11.462757519582956,
-    "gpu_batches_per_second": 1.0508221895136158,
-    "producer_headroom": 10.90837025899559,
+    "producer_batch_seconds": 0.0704138787109656,
+    "producer_p95_seconds": 0.14424603899988142,
+    "gpu_update_seconds": 0.9524318765646669,
+    "producer_batches_per_second": 14.201745711307751,
+    "gpu_batches_per_second": 1.0499438590893313,
+    "producer_headroom": 13.526195318315052,
     "sustainable": true,
-    "buffer_seconds": 60.90469028791929,
+    "buffer_seconds": 60.00320822357401,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 477111.9915279123,
-    "upload_bytes_per_second": 3963640.1632311153,
+    "generation_bytes_per_second": 477172.64382450713,
+    "upload_bytes_per_second": 3934317.189390528,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.37712678316143,
+    "estimated_full_shards_per_hour": 102.39014135409747,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 6009106,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9665561969995906,
+      "bytes": 2658696,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 443100, "loss": 0.6199808120727539, "lr": 0.0003, "grad_norm": 0.14270231877020176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64856751537, "tokens_per_second": 153131.35662853392, "data_wait_s": 1.625300001251162e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443200, "loss": 0.5660464763641357, "lr": 0.0003, "grad_norm": 0.15441132836303142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64871526826, "tokens_per_second": 153809.60971960326, "data_wait_s": 1.765099978001672e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443300, "loss": 0.6141314506530762, "lr": 0.0003, "grad_norm": 0.16403270793021887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64886184359, "tokens_per_second": 152185.15142252238, "data_wait_s": 1.8818999706127215e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 443400, "loss": 0.5952989459037781, "lr": 0.0003, "grad_norm": 0.17503115963409718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64900704977, "tokens_per_second": 151484.64492966785, "data_wait_s": 2.0217000383127015e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2da52a096ecdfa29949311e55fb62616193cb8f698f3407fd3606ad5499749ac
-size 65823
+oid sha256:aa8be554f60e50314cadef19b9d5b7e78a738b54bd71e8c5ad192c1aef2a4366
+size 67941