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": 76973.1499925782,
-      "required_pages_per_second_upper_estimate": 0.005325181118801139
+      "required_token_positions_per_second": 76979.00261472432,
+      "required_pages_per_second_upper_estimate": 0.005325586016781163
     },
     {
       "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": 9621.643749072275,
-      "required_pages_per_second_upper_estimate": 0.004028381314967391
+      "required_token_positions_per_second": 9622.37532684054,
+      "required_pages_per_second_upper_estimate": 0.004028687611301883
     },
     {
       "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": 19243.28749814455,
-      "required_pages_per_second_upper_estimate": 0.004007869306245657
+      "required_token_positions_per_second": 19244.75065368108,
+      "required_pages_per_second_upper_estimate": 0.00400817404295791
     },
     {
       "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": 28864.931247216824,
-      "required_pages_per_second_upper_estimate": 0.001730408554072613
+      "required_token_positions_per_second": 28867.12598052162,
+      "required_pages_per_second_upper_estimate": 0.0017305401249830719
     },
     {
       "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": 28864.931247216824,
-      "required_pages_per_second_upper_estimate": 0.0022441214410564587
+      "required_token_positions_per_second": 28867.12598052162,
+      "required_pages_per_second_upper_estimate": 0.002244292071917295
     },
     {
       "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": 28864.931247216824,
-      "required_pages_per_second_upper_estimate": 0.3702483452908098
+      "required_token_positions_per_second": 28867.12598052162,
+      "required_pages_per_second_upper_estimate": 0.37027649697312276
     }
   ],
   "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": 0.12026150512522804,
-    "producer_p95_seconds": 0.21983899400220253,
-    "gpu_update_seconds": 0.9578404938229617,
-    "producer_batches_per_second": 8.31521274375123,
-    "gpu_batches_per_second": 1.0440151637448214,
-    "producer_headroom": 7.9646474807176615,
+    "producer_batch_seconds": 0.2556272526718715,
+    "producer_p95_seconds": 0.22465072000341024,
+    "gpu_update_seconds": 0.9577676703477777,
+    "producer_batches_per_second": 3.911945966432699,
+    "gpu_batches_per_second": 1.0440945450130794,
+    "producer_headroom": 3.7467353747966317,
     "sustainable": true,
-    "buffer_seconds": 60.34395111084658,
+    "buffer_seconds": 60.33936323190999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396768.56592665875,
-    "upload_bytes_per_second": 4275807.893389077,
+    "generation_bytes_per_second": 399337.4830955931,
+    "upload_bytes_per_second": 4228712.711762951,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.13729794835875,
+    "estimated_full_shards_per_hour": 85.68852777148099,
     "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": 1.0678595420013153,
-      "bytes": 6041574,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7238812670002517,
       "bytes": 11592714,
@@ -562,6 +556,12 @@
       "bytes": 6008739,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3038473750020785,
+      "bytes": 2307657,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -165,3 +165,4 @@
 {"step": 701500, "loss": 0.5888649225234985, "lr": 8e-05, "grad_norm": 0.23307399384670752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102680184060, "tokens_per_second": 152914.71957144633, "data_wait_s": 1.8877999536925927e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701600, "loss": 0.6164185404777527, "lr": 8e-05, "grad_norm": 0.22690946397067135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102694853102, "tokens_per_second": 153106.78212982367, "data_wait_s": 1.749800139805302e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701700, "loss": 0.5553956031799316, "lr": 8e-05, "grad_norm": 0.20827242176660823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102709466655, "tokens_per_second": 152349.58702969798, "data_wait_s": 1.572099790791981e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 701800, "loss": 0.4930936098098755, "lr": 8e-05, "grad_norm": 0.22013987717921163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102723854050, "tokens_per_second": 115924.32924219397, "data_wait_s": 1.7975999071495607e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fb3cec26e8a34488c21735f378156c68e78c56e1c0bd8634c3ed18c3ce4317b
-size 356220
+oid sha256:8b1f576f0f3d99be79bae38d09cf7af0bd9ff87dbb0d3dc58ff1e9191281d7dc
+size 358338