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": 75192.39202405035,
-      "required_pages_per_second_upper_estimate": 0.006517485967269406
+      "required_token_positions_per_second": 74738.12227497603,
+      "required_pages_per_second_upper_estimate": 0.006478111016755797
     },
     {
       "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": 9399.049003006294,
-      "required_pages_per_second_upper_estimate": 0.0037494211756048724
+      "required_token_positions_per_second": 9342.265284372004,
+      "required_pages_per_second_upper_estimate": 0.0037267693012494035
     },
     {
       "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": 18798.098006012588,
-      "required_pages_per_second_upper_estimate": 0.004348609138465532
+      "required_token_positions_per_second": 18684.53056874401,
+      "required_pages_per_second_upper_estimate": 0.0043223373105720775
     },
     {
       "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": 28197.147009018878,
-      "required_pages_per_second_upper_estimate": 0.0017170583138522214
+      "required_token_positions_per_second": 28026.79585311601,
+      "required_pages_per_second_upper_estimate": 0.0017066848222211783
     },
     {
       "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": 28197.147009018878,
-      "required_pages_per_second_upper_estimate": 0.002170694431452917
+      "required_token_positions_per_second": 28026.79585311601,
+      "required_pages_per_second_upper_estimate": 0.0021575803279093335
     },
     {
       "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": 28197.147009018878,
-      "required_pages_per_second_upper_estimate": 0.3501141960716053
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28026.79585311601,
+      "required_pages_per_second_upper_estimate": 0.3479990048439352
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1071812924217852,
-    "producer_p95_seconds": 0.4027518160000909,
-    "gpu_update_seconds": 0.9805247314970116,
-    "producer_batches_per_second": 9.329986394125104,
-    "gpu_batches_per_second": 1.0198620879998148,
-    "producer_headroom": 9.148282403970288,
+    "producer_batch_seconds": 0.1555222331638788,
+    "producer_p95_seconds": 0.43609423100133426,
+    "gpu_update_seconds": 0.9864845109265711,
+    "producer_batches_per_second": 6.429948822470082,
+    "gpu_batches_per_second": 1.013700660196615,
+    "producer_headroom": 6.34304491941728,
     "sustainable": true,
-    "buffer_seconds": 60.792533352814715,
+    "buffer_seconds": 60.17555516652084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336651.9501857303,
-    "upload_bytes_per_second": 5422351.308141541,
+    "generation_bytes_per_second": 337392.52271577355,
+    "upload_bytes_per_second": 5363678.822132393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.23767165354664,
+    "estimated_full_shards_per_hour": 72.39658127884775,
     "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": 1.4190698960010195,
-      "bytes": 13694673,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.8739250739999989,
       "bytes": 5944089,
@@ -562,6 +556,12 @@
       "bytes": 9524825,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1543053290006355,
+      "bytes": 7653463,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 354500, "loss": 0.6181938368827105, "lr": 0.0003, "grad_norm": 0.18626586511146814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51889043122, "tokens_per_second": 148291.947073688, "data_wait_s": 1.5765006537549198e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354600, "loss": 0.5678000226616859, "lr": 0.0003, "grad_norm": 0.1558815531563137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51903795748, "tokens_per_second": 149932.9615391219, "data_wait_s": 1.1664997145999223e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354700, "loss": 0.56468465924263, "lr": 0.0003, "grad_norm": 0.16401846710138815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51918300294, "tokens_per_second": 146860.8932387149, "data_wait_s": 1.1041003745049238e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 354800, "loss": 0.5827666632831097, "lr": 0.0003, "grad_norm": 0.14574788928946014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51932766104, "tokens_per_second": 146994.49737078985, "data_wait_s": 9.767994924914092e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf8d14f9e13c0ae62818225bebc600bd7c61bdcebb40d7f698bbaf7570e50cb5
-size 846492
+oid sha256:47f3ea81f1e613b776f19ba6fc289c5047bb02ec71ba7bc59582550672691d12
+size 848610