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": 75985.25680985807,
-      "required_pages_per_second_upper_estimate": 0.006586209477405791
+      "required_token_positions_per_second": 74378.36429014329,
+      "required_pages_per_second_upper_estimate": 0.006446928106428771
     },
     {
       "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": 9498.157101232258,
-      "required_pages_per_second_upper_estimate": 0.003788956877785327
+      "required_token_positions_per_second": 9297.295536267911,
+      "required_pages_per_second_upper_estimate": 0.00370883019637303
     },
     {
       "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": 18996.314202464517,
-      "required_pages_per_second_upper_estimate": 0.004394462967028775
+      "required_token_positions_per_second": 18594.591072535823,
+      "required_pages_per_second_upper_estimate": 0.004301531391005385
     },
     {
       "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": 28494.471303696777,
-      "required_pages_per_second_upper_estimate": 0.0017351638034580898
+      "required_token_positions_per_second": 27891.886608803732,
+      "required_pages_per_second_upper_estimate": 0.0016984695570566625
     },
     {
       "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": 28494.471303696777,
-      "required_pages_per_second_upper_estimate": 0.0021935832787035466
+      "required_token_positions_per_second": 27891.886608803732,
+      "required_pages_per_second_upper_estimate": 0.002147194640829484
     },
     {
       "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": 28494.471303696777,
-      "required_pages_per_second_upper_estimate": 0.3538059687311022
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27891.886608803732,
+      "required_pages_per_second_upper_estimate": 0.3463238835417725
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07669114355475415,
-    "producer_p95_seconds": 0.21217375900050683,
-    "gpu_update_seconds": 0.9702934897554334,
-    "producer_batches_per_second": 13.039315280075899,
-    "gpu_batches_per_second": 1.0306160049080142,
-    "producer_headroom": 12.65196272712619,
+    "producer_batch_seconds": 0.06335803708590504,
+    "producer_p95_seconds": 0.215937187000236,
+    "gpu_update_seconds": 0.9912560017103055,
+    "producer_batches_per_second": 15.783317255301542,
+    "gpu_batches_per_second": 1.0088211302374035,
+    "producer_headroom": 15.645307956215479,
     "sustainable": true,
-    "buffer_seconds": 60.15819636483687,
+    "buffer_seconds": 60.46661610432864,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 559483.5543777691,
+    "generation_bytes_per_second": 536025.4202030265,
     "upload_bytes_per_second": 5959077.121957105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.02607332944777,
+    "estimated_full_shards_per_hour": 115.0185771424112,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 124900, "loss": 0.6229158714413643, "lr": 0.0003, "grad_norm": 0.17464007645231128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18277769906, "tokens_per_second": 151074.6458839549, "data_wait_s": 1.8365999494562857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125000, "loss": 0.6308581009507179, "lr": 0.0003, "grad_norm": 0.1471120359130923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18292497399, "tokens_per_second": 150460.7163588629, "data_wait_s": 1.7464999473304488e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125100, "loss": 0.6330022923648357, "lr": 0.0003, "grad_norm": 0.15490170281847904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18307168581, "tokens_per_second": 131334.71537567364, "data_wait_s": 1.274699934583623e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 125200, "loss": 0.6066600568592548, "lr": 0.0003, "grad_norm": 0.14261032306092158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18321829653, "tokens_per_second": 149559.8669171572, "data_wait_s": 1.3052000213065185e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f23e3d1c515f3029bb0d11c1678ff12dddd7884302c4e6ea632b6ec54572eb6
-size 215649
+oid sha256:95d6db85fa432e60eb80cf72119c9aac862f6a1b3a333805d49653389ea3891b
+size 217767