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": 77064.45029573975,
-      "required_pages_per_second_upper_estimate": 0.005331497485619746
+      "required_token_positions_per_second": 75054.05778081361,
+      "required_pages_per_second_upper_estimate": 0.0051924138666838435
     },
     {
       "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": 9633.056286967469,
-      "required_pages_per_second_upper_estimate": 0.004033159506263217
+      "required_token_positions_per_second": 9381.757222601702,
+      "required_pages_per_second_upper_estimate": 0.0039279458357344725
     },
     {
       "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": 19266.112573934937,
-      "required_pages_per_second_upper_estimate": 0.004012623167595068
+      "required_token_positions_per_second": 18763.514445203404,
+      "required_pages_per_second_upper_estimate": 0.003907945231784264
     },
     {
       "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": 28899.168860902406,
-      "required_pages_per_second_upper_estimate": 0.0017324610467352544
+      "required_token_positions_per_second": 28145.271667805107,
+      "required_pages_per_second_upper_estimate": 0.0016872660611434325
     },
     {
       "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": 28899.168860902406,
-      "required_pages_per_second_upper_estimate": 0.0022467832649251656
+      "required_token_positions_per_second": 28145.271667805107,
+      "required_pages_per_second_upper_estimate": 0.002188171074205153
     },
     {
       "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": 28899.168860902406,
-      "required_pages_per_second_upper_estimate": 0.370687508637683
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28145.271667805107,
+      "required_pages_per_second_upper_estimate": 0.36101732491636984
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.37292732897654446,
-    "producer_p95_seconds": 0.2370887920005771,
-    "gpu_update_seconds": 0.9567057147240277,
-    "producer_batches_per_second": 2.6814875776049543,
-    "gpu_batches_per_second": 1.0452535033601853,
-    "producer_headroom": 2.5653944894561493,
+    "producer_batch_seconds": 0.3704444543202783,
+    "producer_p95_seconds": 0.2370864430013171,
+    "gpu_update_seconds": 0.9823319641865839,
+    "producer_batches_per_second": 2.699460036012367,
+    "gpu_batches_per_second": 1.0179858097441083,
+    "producer_headroom": 2.651765879419215,
     "sustainable": true,
-    "buffer_seconds": 60.27246002761374,
+    "buffer_seconds": 60.9045817795682,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371686.07824357064,
-    "upload_bytes_per_second": 4707631.099469831,
+    "generation_bytes_per_second": 341135.1421239864,
+    "upload_bytes_per_second": 4722295.145764428,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.75517998199787,
+    "estimated_full_shards_per_hour": 73.19966028012938,
     "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": 0.9569596109995473,
-      "bytes": 5245257,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.285435239000435,
       "bytes": 7473995,
@@ -562,6 +556,12 @@
       "bytes": 6276648,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.272974397001235,
+      "bytes": 8064738,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -175,3 +175,4 @@
 {"step": 457700, "loss": 0.45534321665763855, "lr": 0.0003, "grad_norm": 0.22608440001065322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66993292703, "tokens_per_second": 151032.16086570555, "data_wait_s": 1.5909001376712695e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457800, "loss": 0.5443060398101807, "lr": 0.0003, "grad_norm": 0.17473214796489223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67007858122, "tokens_per_second": 152137.92055715068, "data_wait_s": 1.589800012880005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457900, "loss": 0.6169613003730774, "lr": 0.0003, "grad_norm": 0.1489924384099695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67022638589, "tokens_per_second": 154146.99362011062, "data_wait_s": 8.695998985785991e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 458000, "loss": 0.6119412779808044, "lr": 0.0003, "grad_norm": 0.16279702943387112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67037286837, "tokens_per_second": 151251.47856399947, "data_wait_s": 1.3622000551549718e-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 sha256:9bbb17df542e5a0293276742a65d5bc7d44ec0a4994132e22f3ff3c4a301289d
-size 377400
+oid sha256:623972fe9c67f5fb0c49040ef61881b8b1b03a68a1d4d6014fcb83c5c48d99f1
+size 379518