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": 77171.95588849745,
-      "required_pages_per_second_upper_estimate": 0.0053389349719740655
+      "required_token_positions_per_second": 76966.34088214427,
+      "required_pages_per_second_upper_estimate": 0.005324710048742008
     },
     {
       "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": 9646.494486062182,
-      "required_pages_per_second_upper_estimate": 0.004038785799602666
+      "required_token_positions_per_second": 9620.792610268034,
+      "required_pages_per_second_upper_estimate": 0.004028024960923855
     },
     {
       "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": 19292.988972124363,
-      "required_pages_per_second_upper_estimate": 0.004018220812559642
+      "required_token_positions_per_second": 19241.585220536068,
+      "required_pages_per_second_upper_estimate": 0.00400751476671189
     },
     {
       "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": 28939.483458186543,
-      "required_pages_per_second_upper_estimate": 0.0017348778452856127
+      "required_token_positions_per_second": 28862.3778308041,
+      "required_pages_per_second_upper_estimate": 0.0017302554806575112
     },
     {
       "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": 28939.483458186543,
-      "required_pages_per_second_upper_estimate": 0.0022499175475388343
+      "required_token_positions_per_second": 28862.3778308041,
+      "required_pages_per_second_upper_estimate": 0.0022439229241616615
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28939.483458186543,
-      "required_pages_per_second_upper_estimate": 0.37120462100520185
+      "required_token_positions_per_second": 28862.3778308041,
+      "required_pages_per_second_upper_estimate": 0.37021559280671235
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10575785236750335,
-    "producer_p95_seconds": 0.35906469000474317,
-    "gpu_update_seconds": 0.9553729609565231,
-    "producer_batches_per_second": 9.45556266143765,
-    "gpu_batches_per_second": 1.0467116412827888,
-    "producer_headroom": 9.033588897367629,
+    "producer_batch_seconds": 0.1264412003284292,
+    "producer_p95_seconds": 0.3480586570003652,
+    "gpu_update_seconds": 0.9579252327052546,
+    "producer_batches_per_second": 7.908814511429141,
+    "gpu_batches_per_second": 1.043922809273875,
+    "producer_headroom": 7.576052981283455,
     "sustainable": true,
-    "buffer_seconds": 60.18849654026096,
+    "buffer_seconds": 60.34928966043104,
     "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": 517276.916002403,
-    "upload_bytes_per_second": 5762012.004983836,
+    "generation_bytes_per_second": 486201.8888262812,
+    "upload_bytes_per_second": 5578446.735043732,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.9955845837981,
+    "estimated_full_shards_per_hour": 104.32760714141203,
     "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": 1.0723825640016003,
-      "bytes": 6205162,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3036935650015948,
       "bytes": 11727766,
@@ -562,6 +556,12 @@
       "bytes": 7595959,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 4.429340966999007,
+      "bytes": 11190563,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 598800, "loss": 0.5590025186538696, "lr": 0.0003, "grad_norm": 0.16635564231671893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87646776012, "tokens_per_second": 152676.16736255994, "data_wait_s": 1.6075005987659097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 598900, "loss": 0.6015879511833191, "lr": 0.0003, "grad_norm": 0.17408554008761198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87661158218, "tokens_per_second": 149188.59794796898, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599000, "loss": 0.555098295211792, "lr": 0.0003, "grad_norm": 0.16050930060758728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87675752927, "tokens_per_second": 151080.69406588204, "data_wait_s": 1.4387995179276913e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 599100, "loss": 0.5877408385276794, "lr": 0.0003, "grad_norm": 0.1686098600321335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87690455460, "tokens_per_second": 152727.06789891224, "data_wait_s": 1.0186995496042073e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0935c496f5e0958e546db64642eebe8488e92ef4d097498593a5d74d7020ac7
-size 893871
+oid sha256:835f2c53c7af227768b2c54c4db85e54eb3a6ea3bbb7f85dc9ea322e0a25bbe5
+size 895989