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": 75599.14960961684,
-      "required_pages_per_second_upper_estimate": 0.006552742683868634
+      "required_token_positions_per_second": 74454.29039832999,
+      "required_pages_per_second_upper_estimate": 0.006453509189053433
     },
     {
       "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": 9449.893701202105,
-      "required_pages_per_second_upper_estimate": 0.0037697038859111636
+      "required_token_positions_per_second": 9306.786299791249,
+      "required_pages_per_second_upper_estimate": 0.0037126162038420493
     },
     {
       "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": 18899.78740240421,
-      "required_pages_per_second_upper_estimate": 0.0043721331906484825
+      "required_token_positions_per_second": 18613.572599582498,
+      "required_pages_per_second_upper_estimate": 0.004305922433224167
     },
     {
       "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": 28349.68110360631,
-      "required_pages_per_second_upper_estimate": 0.001726346839928051
+      "required_token_positions_per_second": 27920.35889937375,
+      "required_pages_per_second_upper_estimate": 0.001700203370169813
     },
     {
       "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": 28349.68110360631,
-      "required_pages_per_second_upper_estimate": 0.0021824369282956563
+      "required_token_positions_per_second": 27920.35889937375,
+      "required_pages_per_second_upper_estimate": 0.0021493865165739087
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28349.68110360631,
-      "required_pages_per_second_upper_estimate": 0.35200815902760607
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27920.35889937375,
+      "required_pages_per_second_upper_estimate": 0.34667741410002545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12221520023439325,
-    "producer_p95_seconds": 0.4268946180018247,
-    "gpu_update_seconds": 0.9752490653760104,
-    "producer_batches_per_second": 8.182288275780156,
-    "gpu_batches_per_second": 1.0253790908422422,
-    "producer_headroom": 7.9797689935916845,
+    "producer_batch_seconds": 0.10249432328106423,
+    "producer_p95_seconds": 0.39124788300250657,
+    "gpu_update_seconds": 0.9902451504883822,
+    "producer_batches_per_second": 9.756637909182132,
+    "gpu_batches_per_second": 1.0098509439877659,
+    "producer_headroom": 9.661463374638716,
     "sustainable": true,
-    "buffer_seconds": 60.46544205331264,
+    "buffer_seconds": 60.40495417979132,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340933.01408689393,
-    "upload_bytes_per_second": 4842063.584376955,
+    "generation_bytes_per_second": 328429.3339946892,
+    "upload_bytes_per_second": 4846781.735776416,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.15628830866922,
+    "estimated_full_shards_per_hour": 70.47328963165768,
     "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": 1.954254078998929,
-      "bytes": 8371700,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.207307631000731,
       "bytes": 6036390,
@@ -562,6 +556,12 @@
       "bytes": 12581350,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.4273276979947696,
+      "bytes": 6232998,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -397,3 +397,4 @@
 {"step": 194700, "loss": 0.5743269063532352, "lr": 0.0003, "grad_norm": 0.14870336587591793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28496125885, "tokens_per_second": 131303.34703119096, "data_wait_s": 9.789997420739383e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194800, "loss": 0.4885460287332535, "lr": 0.0003, "grad_norm": 0.15674144452471456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28510869924, "tokens_per_second": 150776.8410855801, "data_wait_s": 8.791997970547527e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194900, "loss": 0.6080953255295753, "lr": 0.0003, "grad_norm": 0.16244985294831388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28525593721, "tokens_per_second": 150274.05671662698, "data_wait_s": 1.1323994840495288e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 195000, "loss": 0.587149441242218, "lr": 0.0003, "grad_norm": 0.14609337480068943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28540343020, "tokens_per_second": 150961.18382806727, "data_wait_s": 1.727200287859887e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67d9448a6b40dee7744c3eb3683c1d58b161925c52b758089e8ac78d99da4d99
-size 850728
+oid sha256:ac038f5ae7ef91d5c7fc4340376ab7948fd43eeb6e9e0603463212796a37270c
+size 852846