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": 75820.79360709172,
-      "required_pages_per_second_upper_estimate": 0.006571954223818181
+      "required_token_positions_per_second": 74477.05392873716,
+      "required_pages_per_second_upper_estimate": 0.006455482274175486
     },
     {
       "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": 9477.599200886465,
-      "required_pages_per_second_upper_estimate": 0.003780756023969389
+      "required_token_positions_per_second": 9309.631741092146,
+      "required_pages_per_second_upper_estimate": 0.0037137512929201156
     },
     {
       "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": 18955.19840177293,
-      "required_pages_per_second_upper_estimate": 0.0043849515501521545
+      "required_token_positions_per_second": 18619.26348218429,
+      "required_pages_per_second_upper_estimate": 0.004307238918757979
     },
     {
       "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": 28432.797602659397,
-      "required_pages_per_second_upper_estimate": 0.0017314081986418155
+      "required_token_positions_per_second": 27928.89522327644,
+      "required_pages_per_second_upper_estimate": 0.0017007231875088564
     },
     {
       "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": 28432.797602659397,
-      "required_pages_per_second_upper_estimate": 0.002188835466473958
+      "required_token_positions_per_second": 27928.89522327644,
+      "required_pages_per_second_upper_estimate": 0.002150043666417995
     },
     {
       "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": 28432.797602659397,
-      "required_pages_per_second_upper_estimate": 0.3530401877728174
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27928.89522327644,
+      "required_pages_per_second_upper_estimate": 0.34678340667365853
     }
   ],
   "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": 0.29574102918761014,
-    "producer_p95_seconds": 0.2861542740029108,
-    "gpu_update_seconds": 0.9723981574508872,
-    "producer_batches_per_second": 3.3813367145808737,
-    "gpu_batches_per_second": 1.0283853299572987,
-    "producer_headroom": 3.2880055909794783,
+    "producer_batch_seconds": 0.1136913666952637,
+    "producer_p95_seconds": 0.2665982350008562,
+    "gpu_update_seconds": 0.9899424871255798,
+    "producer_batches_per_second": 8.79574262380346,
+    "gpu_batches_per_second": 1.0101596941289221,
+    "producer_headroom": 8.70727932912447,
     "sustainable": true,
-    "buffer_seconds": 60.28868576195501,
+    "buffer_seconds": 60.386491714660366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379823.4589270368,
-    "upload_bytes_per_second": 6324460.853954639,
+    "generation_bytes_per_second": 379091.1606027508,
+    "upload_bytes_per_second": 6305030.808584491,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.50127244814232,
+    "estimated_full_shards_per_hour": 81.34413827478306,
     "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": 1.0282262380005704,
-      "bytes": 6456198,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.007465778999176,
       "bytes": 7569670,
@@ -562,6 +556,12 @@
       "bytes": 6032509,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.024946067998826,
+      "bytes": 5141639,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 56600, "loss": 0.6294011455029249, "lr": 0.0003, "grad_norm": 0.13477242866264275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8284336702, "tokens_per_second": 150632.95491428688, "data_wait_s": 1.138600055128336e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 56700, "loss": 0.5716652777045965, "lr": 0.0003, "grad_norm": 0.1419586287745717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8299042034, "tokens_per_second": 150238.0994353629, "data_wait_s": 9.60000033956021e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 56800, "loss": 0.6506004333496094, "lr": 0.0003, "grad_norm": 0.12912725044194975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8313602708, "tokens_per_second": 148084.79006997077, "data_wait_s": 1.1123000149382278e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 56900, "loss": 0.6293017156422138, "lr": 0.0003, "grad_norm": 0.1302298009260594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8328199059, "tokens_per_second": 149435.04813178402, "data_wait_s": 1.0220999683951959e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09da04bb87c426c340a8657cf7ac5c72fa6ef5a9933ec7e4d6de7b92b9dad097
-size 358338
+oid sha256:f614909dd1182e83c636b11036742507a4fae9bc1123a7544e237a42776d56a1
+size 360456