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": 76221.85422151587,
-      "required_pages_per_second_upper_estimate": 0.005273204734112686
+      "required_token_positions_per_second": 76820.65166957182,
+      "required_pages_per_second_upper_estimate": 0.005314630930970698
     },
     {
       "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": 9527.731777689483,
-      "required_pages_per_second_upper_estimate": 0.0039890623336543835
+      "required_token_positions_per_second": 9602.581458696477,
+      "required_pages_per_second_upper_estimate": 0.004020400332052933
     },
     {
       "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": 19055.463555378967,
-      "required_pages_per_second_upper_estimate": 0.003968750532218049
+      "required_token_positions_per_second": 19205.162917392954,
+      "required_pages_per_second_upper_estimate": 0.003999928961487906
     },
     {
       "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": 28583.195333068452,
-      "required_pages_per_second_upper_estimate": 0.0017135189162052504
+      "required_token_positions_per_second": 28807.74437608943,
+      "required_pages_per_second_upper_estimate": 0.001726980288467822
     },
     {
       "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": 28583.195333068452,
-      "required_pages_per_second_upper_estimate": 0.002222217712956744
+      "required_token_positions_per_second": 28807.74437608943,
+      "required_pages_per_second_upper_estimate": 0.002239675413364767
     },
     {
       "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": 28583.195333068452,
-      "required_pages_per_second_upper_estimate": 0.3666345394885706
+      "required_token_positions_per_second": 28807.74437608943,
+      "required_pages_per_second_upper_estimate": 0.3695148135104659
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.5850692206564077,
-    "producer_p95_seconds": 0.34248446700075874,
-    "gpu_update_seconds": 0.9672816379634608,
-    "producer_batches_per_second": 1.7091994668221795,
-    "gpu_batches_per_second": 1.0338250626833208,
-    "producer_headroom": 1.6532772598740315,
+    "producer_batch_seconds": 0.11376027888258022,
+    "producer_p95_seconds": 0.3217081050024717,
+    "gpu_update_seconds": 0.9597419235276183,
+    "producer_batches_per_second": 8.79041445592946,
+    "gpu_batches_per_second": 1.0419467728620309,
+    "producer_headroom": 8.436529278538721,
     "sustainable": true,
-    "buffer_seconds": 60.93874319169803,
+    "buffer_seconds": 60.46374118223996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256136.56212095066,
-    "upload_bytes_per_second": 4028946.77978317,
+    "generation_bytes_per_second": 262354.3180090139,
+    "upload_bytes_per_second": 3987884.1642745095,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.960943677152535,
+    "estimated_full_shards_per_hour": 56.29512934878171,
     "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": 1.333218751999084,
-      "bytes": 7645187,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.3886132959996758,
       "bytes": 6063422,
@@ -562,6 +556,12 @@
       "bytes": 8048155,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 3.3732278300012695,
+      "bytes": 11426245,
+      "pages": 26,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 878100, "loss": 0.5215075016021729, "lr": 3e-05, "grad_norm": 0.30821407539493095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128529281573, "tokens_per_second": 150957.94744914758, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878200, "loss": 0.5371964573860168, "lr": 3e-05, "grad_norm": 0.3090578056224381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128544042853, "tokens_per_second": 139411.23748300635, "data_wait_s": 1.1652999091893435e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878300, "loss": 0.5400705337524414, "lr": 3e-05, "grad_norm": 0.3049035014871074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128558742382, "tokens_per_second": 153393.47326460722, "data_wait_s": 1.1251999239902943e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 878400, "loss": 0.5285022258758545, "lr": 3e-05, "grad_norm": 0.3028094354470684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128573480394, "tokens_per_second": 153739.61314169862, "data_wait_s": 1.648499892326072e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e667b995bf12c482dd116a663ab296f629edff7f95e3bcd30d84f6d02d0d4fc
-size 763890
+oid sha256:9ca5e515163d97f4d44eef07b3fbc6d41c5b9e3c91f032a3c347d5b3ed79ddd7
+size 766008