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": 76451.37406952993,
-      "required_pages_per_second_upper_estimate": 0.005289083449757719
+      "required_token_positions_per_second": 75852.76210778218,
+      "required_pages_per_second_upper_estimate": 0.005247670085272895
     },
     {
       "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": 9556.421758691242,
-      "required_pages_per_second_upper_estimate": 0.004001074229584889
+      "required_token_positions_per_second": 9481.595263472773,
+      "required_pages_per_second_upper_estimate": 0.00396974593859182
     },
     {
       "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": 19112.843517382484,
-      "required_pages_per_second_upper_estimate": 0.003980701265092024
+      "required_token_positions_per_second": 18963.190526945546,
+      "required_pages_per_second_upper_estimate": 0.003949532493798766
     },
     {
       "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": 28669.265276073722,
-      "required_pages_per_second_upper_estimate": 0.0017186786778672228
+      "required_token_positions_per_second": 28444.78579041832,
+      "required_pages_per_second_upper_estimate": 0.0017052214754625095
     },
     {
       "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": 28669.265276073722,
-      "required_pages_per_second_upper_estimate": 0.002228909272443734
+      "required_token_positions_per_second": 28444.78579041832,
+      "required_pages_per_second_upper_estimate": 0.002211456979814934
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28669.265276073722,
-      "required_pages_per_second_upper_estimate": 0.36773855230273755
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28444.78579041832,
+      "required_pages_per_second_upper_estimate": 0.36485917048804295
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3578213551251963,
-    "producer_p95_seconds": 0.24255551200258196,
-    "gpu_update_seconds": 0.9643776962458109,
-    "producer_batches_per_second": 2.7946906624679095,
-    "gpu_batches_per_second": 1.0369381248579905,
-    "producer_headroom": 2.6951373427904817,
+    "producer_batch_seconds": 0.2908263999767655,
+    "producer_p95_seconds": 0.2269909339993319,
+    "gpu_update_seconds": 0.9719883357080257,
+    "producer_batches_per_second": 3.438477387472015,
+    "gpu_batches_per_second": 1.0288189304983477,
+    "producer_headroom": 3.342159913218604,
     "sustainable": true,
-    "buffer_seconds": 60.75579486348609,
+    "buffer_seconds": 60.26327681389759,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413259.03003448294,
-    "upload_bytes_per_second": 4805507.222908407,
+    "generation_bytes_per_second": 391170.84065947135,
+    "upload_bytes_per_second": 4891420.962469844,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.67576766754023,
+    "estimated_full_shards_per_hour": 83.93615641439538,
     "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": 0.9829580149998947,
-      "bytes": 2655368,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.023656208999455,
       "bytes": 5985547,
@@ -562,6 +556,12 @@
       "bytes": 2620310,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1430494310006907,
+      "bytes": 10900083,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 456800, "loss": 0.5900413393974304, "lr": 0.0003, "grad_norm": 0.15120755257622198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66861847308, "tokens_per_second": 151165.7018522815, "data_wait_s": 1.1833999451482669e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456900, "loss": 0.6693919897079468, "lr": 0.0003, "grad_norm": 0.15900929390157792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66876376233, "tokens_per_second": 150931.19403093614, "data_wait_s": 1.1603002349147573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457000, "loss": 0.595064103603363, "lr": 0.0003, "grad_norm": 0.1715079209725015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66890919005, "tokens_per_second": 151269.69374235423, "data_wait_s": 1.1873999028466642e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 457100, "loss": 0.5534655451774597, "lr": 0.0003, "grad_norm": 0.14475705043126316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66905536931, "tokens_per_second": 152039.5822067574, "data_wait_s": 1.2894000974483788e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b23f85d0f9750512987e119f44daa52532889200d257e7bedab4027fed4591ec
-size 358338
+oid sha256:d593faff1280fdbfece19afc93a989b6db42f690c747db7099912dc44b32e265
+size 360456