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": 77258.09606028255,
-      "required_pages_per_second_upper_estimate": 0.005344894348930895
+      "required_token_positions_per_second": 77538.47346836272,
+      "required_pages_per_second_upper_estimate": 0.005364291508587098
     },
     {
       "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": 9657.262007535319,
-      "required_pages_per_second_upper_estimate": 0.004043293935992051
+      "required_token_positions_per_second": 9692.30918354534,
+      "required_pages_per_second_upper_estimate": 0.0040579674567191885
     },
     {
       "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": 19314.524015070638,
-      "required_pages_per_second_upper_estimate": 0.004022705994088078
+      "required_token_positions_per_second": 19384.61836709068,
+      "required_pages_per_second_upper_estimate": 0.004037304799101482
     },
     {
       "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": 28971.786022605957,
-      "required_pages_per_second_upper_estimate": 0.0017368143346993966
+      "required_token_positions_per_second": 29076.92755063602,
+      "required_pages_per_second_upper_estimate": 0.0017431174087629813
     },
     {
       "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": 28971.786022605957,
-      "required_pages_per_second_upper_estimate": 0.0022524289298384787
+      "required_token_positions_per_second": 29076.92755063602,
+      "required_pages_per_second_upper_estimate": 0.0022606032211741093
     },
     {
       "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": 28971.786022605957,
-      "required_pages_per_second_upper_estimate": 0.3716189636177827
+      "required_token_positions_per_second": 29076.92755063602,
+      "required_pages_per_second_upper_estimate": 0.3729676062471751
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11175292648457003,
-    "producer_p95_seconds": 0.2761842220024846,
-    "gpu_update_seconds": 0.9543077523224478,
-    "producer_batches_per_second": 8.94831152487154,
-    "gpu_batches_per_second": 1.047879992137079,
-    "producer_headroom": 8.539443058381215,
+    "producer_batch_seconds": 0.2580700715157036,
+    "producer_p95_seconds": 0.28625006399670383,
+    "gpu_update_seconds": 0.9508569965603273,
+    "producer_batches_per_second": 3.8749165841927153,
+    "gpu_batches_per_second": 1.0516828541173329,
+    "producer_headroom": 3.684491545167288,
     "sustainable": true,
-    "buffer_seconds": 60.12138839631421,
+    "buffer_seconds": 60.85484777986095,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 575952.1371203886,
-    "upload_bytes_per_second": 4842845.143114314,
+    "generation_bytes_per_second": 535018.928310887,
+    "upload_bytes_per_second": 4833850.088496403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.7929605732381,
+    "estimated_full_shards_per_hour": 114.80260741229016,
     "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": 1.156715154000267,
-      "bytes": 10313432,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0891685620008502,
       "bytes": 2577323,
@@ -562,6 +556,12 @@
       "bytes": 11713877,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.506981171998632,
+      "bytes": 11206977,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 918000, "loss": 0.5584174394607544, "lr": 3e-05, "grad_norm": 0.3468850130237999, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134368569035, "tokens_per_second": 153962.1086885535, "data_wait_s": 1.3860000763088465e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 918100, "loss": 0.5291762948036194, "lr": 3e-05, "grad_norm": 0.31930739252881585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134383307774, "tokens_per_second": 153637.1333065604, "data_wait_s": 1.6166999557754025e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 918200, "loss": 0.4973401427268982, "lr": 3e-05, "grad_norm": 0.3176705004802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134397909826, "tokens_per_second": 152576.66192717396, "data_wait_s": 1.6048001270974055e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 918300, "loss": 0.534902811050415, "lr": 3e-05, "grad_norm": 0.36535324151507903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134412416940, "tokens_per_second": 151727.8327362531, "data_wait_s": 1.6460006008856e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4406d7cb41e0e168bf2f8ccb01d1351c5a6fd93d5630cec7f92926572cfbaec1
-size 708039
+oid sha256:934fa8dca8752d1fd6c58697e8972b528aadff92f60cd58515e32d86b062b78a
+size 710157