Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75584.9154966437,
-      "required_pages_per_second_upper_estimate": 0.006551508907032152
+      "required_token_positions_per_second": 75446.23942941042,
+      "required_pages_per_second_upper_estimate": 0.0065394888169956495
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9448.114437080463,
-      "required_pages_per_second_upper_estimate": 0.0037689941108506714
+      "required_token_positions_per_second": 9430.779928676302,
+      "required_pages_per_second_upper_estimate": 0.0037620791162742548
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18896.228874160926,
-      "required_pages_per_second_upper_estimate": 0.004371309987767357
+      "required_token_positions_per_second": 18861.559857352604,
+      "required_pages_per_second_upper_estimate": 0.004363289920882608
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28344.34331124139,
-      "required_pages_per_second_upper_estimate": 0.0017260217963782597
+      "required_token_positions_per_second": 28292.339786028904,
+      "required_pages_per_second_upper_estimate": 0.0017228550545342297
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28344.34331124139,
-      "required_pages_per_second_upper_estimate": 0.00218202601026345
+      "required_token_positions_per_second": 28292.339786028904,
+      "required_pages_per_second_upper_estimate": 0.002178022634937621
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28344.34331124139,
-      "required_pages_per_second_upper_estimate": 0.35194188151087563
+      "required_token_positions_per_second": 28292.339786028904,
+      "required_pages_per_second_upper_estimate": 0.35129617177233946
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07889004692178503,
-    "producer_p95_seconds": 0.20095749800020712,
-    "gpu_update_seconds": 0.9754327237856585,
-    "producer_batches_per_second": 12.675870265249593,
-    "gpu_batches_per_second": 1.0251860283290433,
-    "producer_headroom": 12.364458659186047,
+    "producer_batch_seconds": 0.3826029873123673,
+    "producer_p95_seconds": 0.23842598000010184,
+    "gpu_update_seconds": 0.9772256451427503,
+    "producer_batches_per_second": 2.6136753584298944,
+    "gpu_batches_per_second": 1.0233051137886613,
+    "producer_headroom": 2.554150588335363,
     "sustainable": true,
-    "buffer_seconds": 60.476828874710826,
+    "buffer_seconds": 60.587989998850524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345329.30585986446,
-    "upload_bytes_per_second": 5722520.8860474955,
+    "generation_bytes_per_second": 333852.25946565875,
+    "upload_bytes_per_second": 5744167.496654029,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.099630182714,
+    "estimated_full_shards_per_hour": 71.63692319848367,
     "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": 0.8837165979998645,
-      "bytes": 52118,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 1.4180460719999246,
       "bytes": 7624068,
@@ -562,6 +556,12 @@
       "bytes": 6217801,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5358513270002732,
+      "bytes": 5360685,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -102,3 +102,4 @@
 {"step": 365200, "loss": 0.6003073621541262, "lr": 0.0003, "grad_norm": 0.18070827367960746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53454994492, "tokens_per_second": 147246.25298451452, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365300, "loss": 0.6012932620942593, "lr": 0.0003, "grad_norm": 0.1589549066821825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53469587411, "tokens_per_second": 148940.8781855063, "data_wait_s": 9.033001333591528e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 365400, "loss": 0.5744970254600048, "lr": 0.0003, "grad_norm": 0.1481814624720664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53484263595, "tokens_per_second": 149064.91951471538, "data_wait_s": 1.444200097466819e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 365500, "loss": 0.5785230807960033, "lr": 0.0003, "grad_norm": 0.16260728728899426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53498888747, "tokens_per_second": 149437.7973033231, "data_wait_s": 1.3671999113284983e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3cd96c2e8af8222f3f75094b64407a6d96481e8abb7bdeebe3226d4771d8f72e
-size 222003
+oid sha256:1c952f332b4ca8b8a4edaabc69de7be386d194181af4b2a52265353c63721092
+size 224121