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": 77191.19493726743,
-      "required_pages_per_second_upper_estimate": 0.005340265974008716
+      "required_token_positions_per_second": 76837.94517579672,
+      "required_pages_per_second_upper_estimate": 0.0053158273358577994
     },
     {
       "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": 9648.899367158429,
-      "required_pages_per_second_upper_estimate": 0.004039792673098036
+      "required_token_positions_per_second": 9604.74314697459,
+      "required_pages_per_second_upper_estimate": 0.004021305385793795
     },
     {
       "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": 19297.798734316857,
-      "required_pages_per_second_upper_estimate": 0.0040192225591823795
+      "required_token_positions_per_second": 19209.48629394918,
+      "required_pages_per_second_upper_estimate": 0.00400082940680946
     },
     {
       "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": 28946.698101475286,
-      "required_pages_per_second_upper_estimate": 0.0017353103521346503
+      "required_token_positions_per_second": 28814.22944092377,
+      "required_pages_per_second_upper_estimate": 0.00172736905820252
     },
     {
       "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": 28946.698101475286,
-      "required_pages_per_second_upper_estimate": 0.002250478454320672
+      "required_token_positions_per_second": 28814.22944092377,
+      "required_pages_per_second_upper_estimate": 0.0022401795986308507
     },
     {
       "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": 28946.698101475286,
-      "required_pages_per_second_upper_estimate": 0.3712971627028294
+      "required_token_positions_per_second": 28814.22944092377,
+      "required_pages_per_second_upper_estimate": 0.36959799695904066
     }
   ],
   "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": 0.10905385620338848,
-    "producer_p95_seconds": 0.3308323550008936,
-    "gpu_update_seconds": 0.9551348448474993,
-    "producer_batches_per_second": 9.169781196319846,
-    "gpu_batches_per_second": 1.0469725875822948,
-    "producer_headroom": 8.758377540232473,
+    "producer_batch_seconds": 0.09236757268735118,
+    "producer_p95_seconds": 0.34830649100331357,
+    "gpu_update_seconds": 0.9595259195351787,
+    "producer_batches_per_second": 10.8263102613385,
+    "gpu_batches_per_second": 1.0421813310519303,
+    "producer_headroom": 10.388125308683966,
     "sustainable": true,
-    "buffer_seconds": 60.17349522539246,
+    "buffer_seconds": 60.45013293071626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404343.99085325695,
-    "upload_bytes_per_second": 5866850.4165837895,
+    "generation_bytes_per_second": 416073.32751124154,
+    "upload_bytes_per_second": 5788046.935913467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.76280779073984,
+    "estimated_full_shards_per_hour": 89.27965039256034,
     "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": 1.2149433689992293,
-      "bytes": 11402427,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.7934381430022768,
       "bytes": 2164331,
@@ -562,6 +556,12 @@
       "bytes": 7188821,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.3482005179976113,
+      "bytes": 6310167,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 518900, "loss": 0.6125057339668274, "lr": 0.0003, "grad_norm": 0.1908833882046134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75949320443, "tokens_per_second": 152370.00133606794, "data_wait_s": 1.7129001207649708e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519000, "loss": 0.6022749543190002, "lr": 0.0003, "grad_norm": 0.15604335112727685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75964138195, "tokens_per_second": 154719.1865134, "data_wait_s": 1.1458003427833319e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519100, "loss": 0.6098043322563171, "lr": 0.0003, "grad_norm": 0.16214481521855956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75978985534, "tokens_per_second": 155114.9035491838, "data_wait_s": 1.7710000975057483e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 519200, "loss": 0.5777155160903931, "lr": 0.0003, "grad_norm": 0.16296838538223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75993615308, "tokens_per_second": 152380.59228125404, "data_wait_s": 1.545400300528854e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a9f2f4409ddd9cfbf993a02e91ca6de8989a7d5bab1700da0b0978cf9d5a042
-size 727101
+oid sha256:2d0b0d503b78989c67df3e4476d16aad5e16eb7ad1918effc1ec9aac33da7986
+size 729219