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": 76359.01464298002,
-      "required_pages_per_second_upper_estimate": 0.0052826938103256015
+      "required_token_positions_per_second": 76101.9598717882,
+      "required_pages_per_second_upper_estimate": 0.005264910164805308
     },
     {
       "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": 9544.876830372503,
-      "required_pages_per_second_upper_estimate": 0.003996240609183351
+      "required_token_positions_per_second": 9512.744983973525,
+      "required_pages_per_second_upper_estimate": 0.003982787676085352
     },
     {
       "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": 19089.753660745006,
-      "required_pages_per_second_upper_estimate": 0.003975892256874905
+      "required_token_positions_per_second": 19025.48996794705,
+      "required_pages_per_second_upper_estimate": 0.003962507824412638
     },
     {
       "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": 28634.630491117507,
-      "required_pages_per_second_upper_estimate": 0.0017166023753933523
+      "required_token_positions_per_second": 28538.234951920575,
+      "required_pages_per_second_upper_estimate": 0.0017108236100059622
     },
     {
       "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": 28634.630491117507,
-      "required_pages_per_second_upper_estimate": 0.0022262165702557073
+      "required_token_positions_per_second": 28538.234951920575,
+      "required_pages_per_second_upper_estimate": 0.002218722241082313
     },
     {
       "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": 28634.630491117507,
-      "required_pages_per_second_upper_estimate": 0.36729429446925393
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28538.234951920575,
+      "required_pages_per_second_upper_estimate": 0.36605783599390174
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 720750,
-    "producer_batch_seconds": 0.045587138031226715,
-    "producer_p95_seconds": 0.14222541899971475,
-    "gpu_update_seconds": 0.9655441514629092,
-    "producer_batches_per_second": 21.936011848671228,
-    "gpu_batches_per_second": 1.035685419962294,
-    "producer_headroom": 21.180187946905583,
+    "producer_batch_seconds": 0.06270230299220003,
+    "producer_p95_seconds": 0.15054443400003947,
+    "gpu_update_seconds": 0.9688055356815028,
+    "producer_batches_per_second": 15.948377528085322,
+    "gpu_batches_per_second": 1.0321988914901827,
+    "producer_headroom": 15.450876434347542,
     "sustainable": true,
-    "buffer_seconds": 60.82928154216328,
+    "buffer_seconds": 60.065943212253174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 70,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301380.2446097301,
-    "upload_bytes_per_second": 4714770.321492496,
+    "generation_bytes_per_second": 351814.8630893504,
+    "upload_bytes_per_second": 4963210.500385602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.66918472021987,
+    "estimated_full_shards_per_hour": 75.4912797881163,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 9710127,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.972143672999664,
+      "bytes": 6102938,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 601500, "loss": 0.5248953104019165, "lr": 0.0003, "grad_norm": 0.15204377231219687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88041812100, "tokens_per_second": 152086.48631346013, "data_wait_s": 1.436200000171084e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601600, "loss": 0.5386746525764465, "lr": 0.0003, "grad_norm": 0.1795039055292808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88056375401, "tokens_per_second": 150902.21547646535, "data_wait_s": 1.4191999980539549e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601700, "loss": 0.55489182472229, "lr": 0.0003, "grad_norm": 0.1717567575760197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88070906247, "tokens_per_second": 150671.75487670003, "data_wait_s": 1.571199982208782e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601800, "loss": 0.5576729774475098, "lr": 0.0003, "grad_norm": 0.14129920849856534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88085408367, "tokens_per_second": 150070.21395653248, "data_wait_s": 1.1115999768662732e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47cf2cf9323970ede89754ce0cf257a6322d5a5af1eea8311ede74f4f3f4b8fe
-size 36171
+oid sha256:c1893723d15a0c415292877dee0ddb3a9d15eb80605ccb069dbdc3988211f51a
+size 38289