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": 76750.9427922613,
-      "required_pages_per_second_upper_estimate": 0.0053098083090914764
+      "required_token_positions_per_second": 76770.06155281387,
+      "required_pages_per_second_upper_estimate": 0.0053111309893081405
     },
     {
       "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": 9593.867849032662,
-      "required_pages_per_second_upper_estimate": 0.004016752125647555
+      "required_token_positions_per_second": 9596.257694101734,
+      "required_pages_per_second_upper_estimate": 0.004017752703872336
     },
     {
       "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": 19187.735698065324,
-      "required_pages_per_second_upper_estimate": 0.00399629933128864
+      "required_token_positions_per_second": 19192.51538820347,
+      "required_pages_per_second_upper_estimate": 0.00399729481469551
     },
     {
       "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": 28781.603547097984,
-      "required_pages_per_second_upper_estimate": 0.0017254131856845195
+      "required_token_positions_per_second": 28788.7730823052,
+      "required_pages_per_second_upper_estimate": 0.0017258429883729458
     },
     {
       "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": 28781.603547097984,
-      "required_pages_per_second_upper_estimate": 0.0022376430788920367
+      "required_token_positions_per_second": 28788.7730823052,
+      "required_pages_per_second_upper_estimate": 0.002238200478718945
     },
     {
       "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": 28781.603547097984,
-      "required_pages_per_second_upper_estimate": 0.369179507023999
+      "required_token_positions_per_second": 28788.7730823052,
+      "required_pages_per_second_upper_estimate": 0.36927147012359
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.20074863879705163,
-    "producer_p95_seconds": 0.29929958100183285,
-    "gpu_update_seconds": 0.9606136070478851,
-    "producer_batches_per_second": 4.98135382631888,
-    "gpu_batches_per_second": 1.0410012857023287,
-    "producer_headroom": 4.785156267081964,
+    "producer_batch_seconds": 0.10703674067951852,
+    "producer_p95_seconds": 0.2986441940010991,
+    "gpu_update_seconds": 0.9603743765306078,
+    "producer_batches_per_second": 9.34258642080784,
+    "gpu_batches_per_second": 1.0412606004884692,
+    "producer_headroom": 8.972380609066652,
     "sustainable": true,
-    "buffer_seconds": 60.518657244016765,
+    "buffer_seconds": 60.50358572142829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481386.61246356735,
-    "upload_bytes_per_second": 5806526.815700199,
+    "generation_bytes_per_second": 498051.040233877,
+    "upload_bytes_per_second": 5729786.077755041,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.29436092787041,
+    "estimated_full_shards_per_hour": 106.87015919935449,
     "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": 1.2800072050013114,
-      "bytes": 12000660,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 1.2026952199994412,
       "bytes": 6686271,
@@ -562,6 +556,12 @@
       "bytes": 10179142,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.5255058740003733,
+      "bytes": 7690292,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 587000, "loss": 0.5778745412826538, "lr": 0.0003, "grad_norm": 0.16915777713992394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85919437460, "tokens_per_second": 152308.90628709272, "data_wait_s": 1.4666999049950391e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587100, "loss": 0.5684595704078674, "lr": 0.0003, "grad_norm": 0.17902361592404403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85934066337, "tokens_per_second": 152722.94810485, "data_wait_s": 1.0110998118761927e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587200, "loss": 0.5829356908798218, "lr": 0.0003, "grad_norm": 0.17970318356402584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85948598787, "tokens_per_second": 150609.5345376927, "data_wait_s": 1.430700285709463e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 587300, "loss": 0.526847779750824, "lr": 0.0003, "grad_norm": 0.1817861976746766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85963125034, "tokens_per_second": 151215.9652437424, "data_wait_s": 2.407300053164363e-05, "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 sha256:92051790beb078ed7368c510b5b38a2953caa7dab6557efa301ed2bbc45f97f2
-size 642381
+oid sha256:824648bfe81c8314d4b3acb6400b27159c228219d7a1bda0903e8b16199aa265
+size 644499