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": 75658.01733041806,
-      "required_pages_per_second_upper_estimate": 0.0065578451887088095
+      "required_token_positions_per_second": 75498.9219884817,
+      "required_pages_per_second_upper_estimate": 0.006544055207693223
     },
     {
       "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": 9457.252166302258,
-      "required_pages_per_second_upper_estimate": 0.003772639287658472
+      "required_token_positions_per_second": 9437.365248560212,
+      "required_pages_per_second_upper_estimate": 0.003764706098835253
     },
     {
       "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": 18914.504332604516,
-      "required_pages_per_second_upper_estimate": 0.004375537693441198
+      "required_token_positions_per_second": 18874.730497120425,
+      "required_pages_per_second_upper_estimate": 0.004366336716597551
     },
     {
       "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": 28371.756498906772,
-      "required_pages_per_second_upper_estimate": 0.0017276911156812011
+      "required_token_positions_per_second": 28312.095745680635,
+      "required_pages_per_second_upper_estimate": 0.0017240580888255104
     },
     {
       "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": 28371.756498906772,
-      "required_pages_per_second_upper_estimate": 0.002184136353334491
+      "required_token_positions_per_second": 28312.095745680635,
+      "required_pages_per_second_upper_estimate": 0.002179543503399608
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28371.756498906772,
-      "required_pages_per_second_upper_estimate": 0.3522822615556424
+      "required_token_positions_per_second": 28312.095745680635,
+      "required_pages_per_second_upper_estimate": 0.3515414746722703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10019923403904585,
-    "producer_p95_seconds": 0.18058466900038184,
-    "gpu_update_seconds": 0.9744902470548603,
-    "producer_batches_per_second": 9.980116211370618,
-    "gpu_batches_per_second": 1.026177535406061,
-    "producer_headroom": 9.725525912454769,
+    "producer_batch_seconds": 0.4158118596484712,
+    "producer_p95_seconds": 0.1776672599999074,
+    "gpu_update_seconds": 0.97654374470735,
+    "producer_batches_per_second": 2.404933810318454,
+    "gpu_batches_per_second": 1.0240196667274535,
+    "producer_headroom": 2.3485230689016983,
     "sustainable": true,
-    "buffer_seconds": 60.41839531740134,
+    "buffer_seconds": 60.5457121718557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474203.5777259093,
-    "upload_bytes_per_second": 5784135.750922963,
+    "generation_bytes_per_second": 454212.6587286632,
+    "upload_bytes_per_second": 5864309.12041668,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.75304888565978,
+    "estimated_full_shards_per_hour": 97.46346303362773,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 7504983,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2214578000002803,
+      "bytes": 10891364,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 86900, "loss": 0.5962490197271109, "lr": 0.0003, "grad_norm": 0.1517551616237985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12717146045, "tokens_per_second": 147696.89799948194, "data_wait_s": 1.0507999832043424e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 87000, "loss": 0.6176202818751335, "lr": 0.0003, "grad_norm": 0.12908852094214038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12731783694, "tokens_per_second": 148367.29145016827, "data_wait_s": 1.1268999514868483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 87100, "loss": 0.6127027161419392, "lr": 0.0003, "grad_norm": 0.13694301061533232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12746455790, "tokens_per_second": 148146.26010645452, "data_wait_s": 1.1270999493717682e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 87200, "loss": 0.5842791199684143, "lr": 0.0003, "grad_norm": 0.13671687230815713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12761166726, "tokens_per_second": 149832.0429945366, "data_wait_s": 1.1705999895639252e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7dc166f2a058bf2ab69363a9b40cb6fb10a148190362bc340142fa085deeecc
-size 111084
+oid sha256:b0136755fbb9bea884e35b4688e31458683411ff31d7acddf73fb6d48f10031c
+size 113202